7f353b73 reworded ValidateDownloadURL's rejection from "下载地址必须是
受信任域名上的 HTTPS URL" to "下载地址必须是合法的 HTTPS URL" when
the IP-literal refusal was removed, but missed that the string is a
test contract: paramAliasExpectedCaptureBoundaryError treats the URL
validation stop as the expected capture boundary for drive +download
and +version-download fixture runs. The stale assertion failed those
subtests and the alias-count invariant (58 active, want 64), which
also failed the Coverage jobs running the same tests.
- update the boundary matcher to the new message
Confirmed with the product team that the official GUI client applies no
client-side SSRF interception to downloads, so the IP-literal refusal
was the last remaining client-side interception beyond transport
hygiene. Dedicated deployments make every host dimension (domain,
port, network location) unenumerable, and an IP-literal host is just
another network location.
- ValidateDownloadURL accepts any HTTPS host, IP literals included;
HTTPS scheme, userinfo refusal, and per-hop redirect re-validation
stay as the transport baseline
- this retires the third and last interception layer after the host
allowlist removal and the dial-time public-IP refusal (be52ca5d)
- upload targets stay unaffected: trustedUploadHost keeps rejecting
non-DingTalk/OSS hosts, so local file bytes cannot be PUT to an IP
- regressions: IP-literal URLs pass validation on the shared and chat
download paths; userinfo and plain-HTTP URLs stay rejected
Customer round-2 testing on the dedicated deployment (Jingbo) found the
storage domain resolving to a customer-intranet address (10.254.87.52),
which the dial-time public-IP policy refused: dedicated storage can be
deployed inside the customer network, so its resolved network location
is as unenumerable as its domain and port.
- delete the public-IP policy, the IANA special-purpose denylist, and
the NAT64 embedded-IPv4 re-validation introduced in 617b780a;
downloads now dial the service-issued host directly, still ignoring
environment proxies
- align with the official GUI client, which applies no client-side
SSRF interception to downloads: no command accepts a user-supplied
download URL, TLS hostname verification pins the connection to the
requested domain, redirects are re-validated per hop, and credential
headers are stripped once a redirect leaves the original origin
- uploads keep the static DingTalk/OSS default-port trust boundary
- simplify SetSecureDownloadDialTargetForTest to a single dial seam
Review finding on 1ba6bec8: relaxing ValidateDownloadURL to accept
non-default HTTPS ports also widened upload targets, because the upload
validator reuses it and only re-imposed the host trust set.
- validateUploadURL now also rejects non-default ports, making the
upload trust boundary identical to the pre-removal policy (trusted
DingTalk/OSS hosts on the default port)
- DingTalk/OSS upload endpoints always serve HTTPS on 443, so unlike
dedicated-deployment downloads there is no legitimate non-default
port scenario for uploads
- regressions: trusted-host:8443 upload targets are rejected for both
public-cloud and dedicated hosts
Customer testing on a dedicated deployment found real download URLs served
on a non-default HTTPS port (e.g. 8443) by the dedicated storage domain,
which the inherited default-port-only rule rejected.
- drop the 443-only restriction from ValidateDownloadURL; HTTPS scheme,
domain-only hosts (no IP literals), and no-userinfo rules stay
- the port is not a trust signal: SSRF protection lives at dial time in
the port-agnostic public-IP policy
- redirect hygiene unchanged: a port change is a cross-origin redirect
and still strips service credential headers (new regression guard)
- dedicated-deployment regression: same host on a non-default port is
accepted by URL validation and downloads successfully
- re-validate NAT64 well-known prefix answers (64:ff9b::/96) against the
embedded IPv4 address: DNS64-synthesized answers for public IPv4-only
hosts keep working while embedded loopback/private/special addresses
are refused before dialing
- refuse NAT64 local-use (64:ff9b:1::/48) outright: the IPv4 embedding
is deployment-specific and cannot be extracted reliably
- refuse Teredo (2001::/32) outright as part of the transition-mechanism
audit; 6to4 (2002::/16) was already refused and IPv4-mapped addresses
are normalized via Unmap before checks
- add dial-layer regression: a hostile AAAA answer embedding 127.0.0.1
fails before any dial attempt
- restore the pre-existing DingTalk/OSS trusted host requirement for
upload target URLs via a dedicated upload validator
- extend the dial-time non-public IP denylist with IANA special-purpose
ranges (0.0.0.0/8, 192.88.99.0/24, 100::/64, 2002::/16, 3fff::/20,
5f00::/16)
- document that download credential headers follow the service-issued
URL as-is on the first hop (same authenticated response issues both);
cross-host redirects keep stripping them
Bind coverage repair to the stable PR head and merge identity plus protected-main containment without treating the live base SHA projection as permanent identity.
Add semantic regression coverage for base advancement and update the governance documentation.
GitHub omits merge-related repository settings from tokens without Contents write. Accept only the exact dual omission in read-only admission, and require the dedicated App to observe the reviewed values before any auto-merge mutation.
Reject report submissions without visible recipients at the CLI layer
(Cobra required flag + fail-closed on blank values) while keeping the
openAPI create_report parameter optional (bug 85724185).
- drop shareScope.required=["linkShare"] from the result schema: the
production response omits linkShare unless link sharing is configured;
document the conditional presence in the schema description, tests,
and both skill references
- spell out policy value tiers in skill docs (READER_AND_ABOVE /
DOWNLOADER_AND_ABOVE / EDITOR_AND_ABOVE / MANAGER_AND_ABOVE) so the
shorthand no longer diverges from the schema enums
Sync the get-setting schema descriptions and skill references with the
server-side finalized policy texts: policy code semantics now quote the
product permission-settings page labels (e.g. external_share=add
external collaborators, watermark=show watermark, node_spread_scope=
download and distribution scope), the node_spread_scope binary values
read ALL_NODES=all documents and PREVIEWABLE_ONLY=previewable documents
only (online documents, images, videos, etc.), and the contract test
value-semantic fragments follow the same wording. Both mono and multi
drive references also list per-policy name/description examples
(external_share, node_spread with NOBODY, node_move_forbidden) quoted
verbatim from the server i18n single source.
Sync the get_permission_setting result contract with the finalized
server-side enums: switch values ON/OFF become ENABLED/DISABLED and the
node_spread_scope binary ALL_CONTENT becomes ALL_NODES (restriction
applies to all nodes vs previewable-only nodes), while policy codes,
threshold domains, permissionMode and shareScope stay unchanged.
Policies now carry per-entry name and description fields (Chinese label
and value semantics from the server i18n single source) as deterministic
required members, mirrored in the contract test assertions and both mono
and multi drive skill quick references.
Declare the drive.permission get_setting ResultSpec data schema from the
server-side output metadata so the field-level contract (permissionMode,
shareScope, policy codes and their typed value domains, inherited /
allowedValues semantics) is reviewed code guarded by contract tests, and
add field-quick-reference rows for the get-setting section in both mono
and multi drive skill references.
Auto-CR flagged that the destructive-command tests only checked the error
string, so a regression that let a command slip past the confirmation gate
(or dispatch the wrong payload) would go unnoticed.
Every user_required leaf across college-contact (15), edu-contact (2),
edu-group (1) and edu-app (3) is now verified in pairs with a fresh
non-dry-run capture caller: without explicit confirmation the gate must
return confirmation_required AND the caller must see zero calls; with
--yes the command must produce exactly one call carrying the correct
productID, tool name and complete argument payload.
Remove custom collegeContactConfirmDestructive function and rely on the
standard framework Confirmation: "user_required" in LeafSpec. This ensures
destructive commands return the proper category:validation/code:3 error
instead of category:internal/code:5.
Also adds run_all_edu_commands_test.go covering all 153 leaf commands
across 5 edu/college products.
The command matrix test with 5 additional edu products (153 leaf commands)
exceeds the 10-minute go test timeout on Windows CI runners. The dedicated
TestCrossPlatformCoverage* tests in individual edu test files already
provide 100% changed-code coverage without the matrix.
The CI platform coverage gate (macOS/Windows) only executes tests
matching ^(TestAllShortcuts|TestCrossPlatformCoverage). The edu/college
test functions used standard names and were not exercised during the
platform coverage run, causing 73.89% changed-code coverage (target 100%).
Changes:
- Rename all edu/college test functions with TestCrossPlatformCoverage prefix
- Add edu products to command_matrix_test.go selected map
- Add TestCrossPlatformCoverageOpenSupplementServersIncludesEduEndpoints
in pkg/edition to cover the new supplement server entries
Local verification: coverage-gate-platform reports 100.0000% (3336 stmts).
The schema catalog validator requires Examples to demonstrate actual
execution, not just --help. Replace all 65 occurrences with realistic
parameter examples derived from each command's required flags.
- schemaPublishedShortcutCount: 461→468 after merging aisearch/contact/live
shortcuts from upstream/main
- Fix P2: datasource-update --field-ids doc says "不传时同步全部字段" but
actual behavior keeps existing field config; fixed in usage guide and
reference
Rebuild the exact-entry reviewedCompatibilityExceptions carve-out in the
base-owned schema-compat checker for the three destructive batch-remove
tools whose confirmation PR #1085 tightens from not_required to
user_required (doc/doc.remove_permission, drive/drive.permission_remove,
wiki/wiki.remove_member). Because the compatibility gate builds its
checker from the PR merge-base, this carve-out has to land on main before
PR #1085 can pass; the entry set is exact (tool + field + old -> new), so
any other confirmation drift, including weakening a reviewed tool back to
not_required, still fails.
Address the P1 review finding on PR #1085: --members lets one call remove
up to 30 USER/DEPT/CONVERSATION/TAG members, where departments, chats,
and role groups can indirectly affect many more users, yet the remove
branches called the MCP tool right after argument parsing with Safety
confirmation=not_required.
- drive permission remove, doc permission remove, and wiki member remove
now declare confirmation=user_required. DeclareLeafMetadata installs
the ConfirmSafety gate automatically (deferred to the first
deps.Caller.CallTool so flag validation still fails first), so an
unconfirmed invocation exits with the typed confirmation_required
error and performs zero MCP calls; --yes, an interactive yes, or
--dry-run previews remain the supported paths.
- Pass framework confirmation errors through WrapErrorWithOperation
verbatim (new apperrors.IsConfirmationRequired). Text classification
misrouted them: command paths containing "permission" (drive/doc
permission remove) were re-reported as AUTH_PERMISSION_DENIED while
other paths (wiki member remove) lost their reason and degraded to
UNCLASSIFIED.
- Tests: TestPermissionMemberRemoveRequiresConfirmationBeforeToolCall
covers all three entry points for both --members and legacy --users —
unconfirmed rejects with zero MCP calls, --yes dispatches exactly one
call with the complete precise arguments, --dry-run previews without
calls. Existing remove tests inject root --yes for the assembly
assertions; blank --users still fails validation before confirmation.
Address the P2 review finding on PR #1085: NO_PERMISSION is a generic
code name also returned by non-document tools — attendance
get-self-setting (bossAttendStatNotify) and event-subscription attempts
have both been observed returning it — so keying drive permission
apply-* guidance on it would mislead those products, defeating the goal
of the P1 scoping fix. Only the drive-specific forbidden.* domain codes
(forbidden.no.auth / forbidden.accessDenied) and the role-threshold
message wording remain document signals; a bare NO_PERMISSION still
classifies as AUTH_PERMISSION_DENIED but now keeps the product-neutral
suggestion, and NO_PERMISSION combined with document wording still gets
apply guidance.
Add regression tests for the non-document NO_PERMISSION case and update
the changelog fragment; changed-code coverage stays at 100%.
Address the two P1 review findings on PR #1085:
- Permission suggestions: the drive permission apply-* guidance is now
limited to document/wiki-specific errors (node access codes
NO_PERMISSION / forbidden.no.auth / forbidden.accessDenied and the
role-threshold wording). Permission failures from other products keep
their product-specific suggestion (e.g. the mail mailbox hint) or fall
back to a product-neutral hint instead of being told to run document
permission commands that cannot fix their problem.
- Null rendering: the null->{} adaptation is limited to the four tools
with a confirmed empty-response-means-success contract
(update_permission / remove_permission / update_member /
remove_member). Every other tool keeps its raw null output so the
shared machine-output contract stays unchanged.
Update tests and the changelog fragment accordingly; changed-code
coverage stays at 100%.
The server rejects the legacy maxResults path; the CLI now validates
--limit (1-50) and sends it as pageSize at runtime. Schema-compat
rejects a non-empty property redirect (maxResults -> pageSize), so
declare --limit as a CLI pagination input via the reviewed mapping
exclusion ledger (property omitted, provenance
reviewed_mapping_exclusion) on doc.list_permission,
drive.list_permission, and wiki.list_member.
- --notify now defaults to false and is omitted from the server request
unless passed explicitly (help updated accordingly)
- forbidden.accessDenied / permission-denied bodies classify as
AUTH_PERMISSION_DENIED with apply-permission guidance
- user/member validation failures intercepted before RESOURCE_NOT_FOUND
with --members corpId suggestion
- business error display appends backend code/logId for traceability;
literal null tool responses render as {}
- drive/doc permission list + wiki member list declare cursor pagination
(next-token) in Contract; cobra.NoArgs hardening on permission leaves
- cross-platform coverage tests and release fragments updated
- Add trimNonEmpty for --table-ids in +datasource-sync and --task-ids in
+datasource-sync-status, matching the existing field-ids pattern
- Add 4 test cases: whitespace-only rejection and trim-through for both
- Fix usage guide: typical workflow and notes no longer equate result
with processCode; correctly describe result as JSON to parse for
approvals[].processCode/name/iconUrl/url
--field-ids is declared as FlagStringSlice, but DatasourceCreate and
DatasourceUpdate previously called rt.Str to check whether the flag
was empty. RuntimeContext.Str delegates to cobra's GetString, which
returns an empty string on slice-typed flags, so the empty-value
guard rejected every explicit --field-ids input and the downstream
MCP tool never received fieldIds.
Switch to rt.StrSlice, sanitize through a new trimNonEmpty helper
(drop whitespace-only / empty entries) and pass the cleaned slice
to MCP. Add success-passthrough tests for both create and update,
plus a whitespace-only rejection case, and enhance the mock caller
to record MCP arguments so fieldIds can be asserted.
- Update --field-ids description in create/update shortcuts and the
helper-layer datasource update to clarify that omitting the flag
keeps existing config (create defaults to all fields), matching the
actual update overwrite semantics.
- Rename datasource shortcut coverage tests to the
TestCrossPlatformCoverage* prefix so they are picked up by the
macOS platform coverage gate.
Align shortcut layer validation with helper layer to prevent empty slices
from being sent to MCP, which could clear sync field selection due to
datasource update's overwrite semantics.
- Add empty string checks for --field-ids in both create and update shortcuts
- Add empty string checks for --auto-sync-setting in both create and update shortcuts
- Add regression tests verifying MCP is not called when empty values are rejected
- Both public entry points now have consistent validation behavior
Fixes P1 auto-CR issue for empty flag bypass vulnerability.
Make +datasource-sync-status consistent across shortcut and native
commands: --task-ids is now required, descriptions focus on querying
by taskId, and optional/IDLE semantics are removed. Update tests,
usage guide, reference doc, and SKILL description accordingly.
Native datasource create/update now expose --field-ids and
--auto-sync-setting, matching the shortcut-layer capabilities:
- flags registered on both commands
- Contract Parameters updated
- values mapped to MCP tool args
- JSON validation for --auto-sync-setting
- no-change update guard now counts the new flags
Also fixes the missing required name in the usage-guide update example.
+datasource-update now requires at least one mutable option
(--source-config, --auto, --field-ids, or --auto-sync-setting)
before calling update_datasource_config, preventing accidental
sync triggers. The native datasource update command enforces the
same guard for its supported flags. Also adds the required name
field to the +datasource-get-fields doc example.
Omitting --auto on +datasource-update previously sent auto=false to
MCP, silently disabling auto-sync for existing datasources. Now auto
is only included in tool args when the flag is explicitly provided,
so --auto=true and --auto=false work while omission preserves the
existing setting. Updated flag descriptions and added tests.
MCP requires the auto field in create_datasource / update_datasource_config
requests. Previously CLI only sent it when --auto was explicitly changed,
causing failures when users omitted the flag. Now both shortcut and helper
layers always include auto=false by default.
Also update flag descriptions and docs to clarify that the field is always
sent downstream, and add test assertions for the default-false behavior.
- docs/datasource-usage-guide.md: clarify that list-sources result is a
JSON string containing approvals[]; add missing --auto-sync-setting
parameter table rows and a dedicated autoSyncSetting format section
using the correct scheduled/daily/weekly/monthly enums.
- skills/references/aitable/aitable-datasource.md: fix autoSyncSetting
enums (schedule/day/week/month -> scheduled/daily/weekly/monthly) and
update the create example accordingly.
The OA approval source-config contract requires processCode, name,
iconUrl, and url to be passed through unchanged from +datasource-list-sources.
Published examples for +datasource-create, +datasource-update, and
+datasource-get-fields were missing `name`, and the usage guide marked it
as optional. Fix all examples in the shortcut layer, helper layer, and
docs; update flag descriptions to mention name; and add a contract test
that validates every delivered example's source-config JSON contains the
required members.
Both the shortcut (+datasource-sync, +datasource-sync-status) and
helper (datasource sync, datasource sync-status) layers now validate
that table-ids contains 1-5 IDs and task-ids contains at most 5 IDs
before calling MCP, matching the declared contract.
ValidateRequiredFlags calls GetString which returns empty for
StringSlice flags, causing the examples test to report --table-ids
as missing. Switch to String + parseCSVValues to match the codebase
convention used by record-ids and other comma-separated flags.
- Add 7 datasource leaf commands to internal/helpers/aitable.go so
coverage test can find tool name literals (fixes TestAllShortcutsAssemble)
- Add 7 entries to semantic_catalog_aitable.json and update catalog count
from 93 to 100 (fixes TestCrossPlatformCoverageAITableSemanticCatalog)
- Update publicShortcutCount/schemaPublishedShortcutCount/publiclyDelivered
from 422/447/422 to 429/454/429 (fixes TestDeliverySchemaCoversOrExactly)
- Fix Contract.Selection.AgentSummary and UseWhen[0] in datasource.go to
match Description and Intent exactly as required by schema contract test
- Fix autoSyncSetting enum: scheduled/daily/weekly/monthly; mark
selectedMonthDays/selectedWeekdays as required for monthly/weekly
- Remove splitParentTableField from --source-config user-settable fields;
add note that splitParentTableField/enableDataSyncOaDetailList are
internal downstream fields not to be passed
- Prepend sync-is-fire-and-forget notice to DatasourceSync descriptions
- Remove --conflict-strategy flag (syncConflictStrategy not in MCP schema)
- Add --auto-sync-setting flag to DatasourceCreate (was only in Execute, not in Flags)
- Expand DatasourceSync description: add 文档链接, errorCode=4014 幂等冲突, 非数据源表参数错误
- Simplify DatasourceGetFields description: remove field property enumeration to match snapshot
- Add --auto-sync-setting flag (JSON string) to +datasource-create and
+datasource-update, validated and passed through as raw string.
- Update +datasource-update --source-config desc to reflect full
replacement semantics ("传入时整体覆盖") and spell out required /
optional fields with defaults.
- Append "仅支持 OA 审批数据源 (datasourceType=OA)" to
+datasource-get-config description.
- Simplify +datasource-list-sources / +datasource-get-fields
descriptions to concise Chinese aligned with snapshot wording.
- Update SKILL.md shortcuts table and add datasource usage guide.
Add 5 data source sync management shortcuts to the aitable service:
- +datasource-create (create_datasource): create sync config + first sync
- +datasource-update (update_datasource_config): update existing sync config
- +datasource-sync (run_datasource_sync): trigger manual sync (max 5 tables)
- +datasource-sync-status (get_datasource_sync_status): query sync task status
- +datasource-get-config (get_datasource_config): get sync config details
Each shortcut declares a full Contract (Identity/Interface/Selection),
Safety, Flags, and Execute that calls rt.CallMCPData on the "aitable"
MCP server. datasource-type is passed through without CLI enum check;
source-config is validated as a JSON object via parseJSONObject.
Sync the permission CRUD overhaul from the internal CLI (MR 28965577):
- drive/doc permission list and wiki member list now accept --next-token
to follow the server cursor (totalCount/hasMore/nextToken); --limit maps
to pageSize capped at 50 instead of the rejected maxResults=200 path
(fixes#1065)
- permission add/update/remove and wiki member add/update/remove accept a
--members JSON array (USER/DEPT/CONVERSATION/TAG grantee types, each with
its own roleId) with optional --notify; legacy --users/--role stays
- cursor/page-token hidden cross-product aliases now resolve to next-token
- regenerate param_aliases_generated.go; wiki member list override no
longer blocks cursor
- update mono/multi skill references and add change fragment
A stamp-shaped directory name is not ownership proof: pruneSkillBackups
counted and RemoveAll'd any 20260819-120000-shaped entry under
~/.dws/skill-backups, so a user or tool that created such a directory
lost its contents once DWS held five backups, and the Go backup path
(MkdirAll) adopted a same-named foreign root outright. The PowerShell
installers already implemented the correct contract; every other
surface now matches it.
Go stamps a freshly created root with the exact marker bytes the
install scripts write (.dws-skill-backup = "dws skill backup v1")
before any payload moves in, claims the root with mkdir so an existing
unproven root bumps to a collision suffix instead of being adopted,
and prunes only roots whose marker verifies — unmarked or wrongly
worded stamp-shaped directories are foreign data, preserved and never
counted against the keep limit. The shell installers (install.sh,
install-skills.sh, install-event.sh, install-devapp.sh) and the npm
installer apply the same rule in their backup collision loops, with
roots recorded as created by the running process exempt from marker
re-verification so a mid-run marker permission failure still reuses
this run's own root and keeps the sibling payload intact.
Regression tests cover every surface: pruning an unmarked/wrongly
marked stamp-shaped directory alongside marked ones, refusing to adopt
a foreign root (payload moves to a suffixed root, foreign data and its
nonexistent marker untouched), same-stamp reuse of a proven root, and
marker-write failure cleaning the empty fresh root.
copy_tree published staged children with mv, which replaces a
concurrently created same-name directory (POSIX rename succeeds over an
empty target) and whose rollback moved every dest child back — including
a concurrent writer's different-named entries — before deleting the
staging tree. Children now publish through kernel-level no-clobber
primitives (mkdir claim + recursion for directories with the recorded
mode restored, ln for regular files, ln -s for symlinks), a manifest
records exactly what this transaction published, the rollback retracts
only those entries in reverse order, and each level re-counts the
destination so a foreign different-named entry aborts the publish with
the destination retained. Read-only staged directories (0555 skill
trees) are made owner-writable for the move; the backup restore uses the
same discipline so a concurrent writer is refused without partially
draining the backup.
Regression tests cover both scripts: a concurrently created same-name
empty child directory and a different-named foreign entry mid-publish
are retained with the original backup kept; both fail against the
previous mv-based implementation.
The platform coverage gates execute only TestCrossPlatformCoverage-named
tests, so the child-move error and dispatch branches that the full local
suite covered incidentally were reported as uncovered changed code on
Windows (96.78% vs the 100% target). Adds a seam-driven edge suite for
the child-move fallback — source/claim/child stat and read failures,
per-child link and symlink collisions and publish failures, rollback
rename failure, foreign-entry abort, mode-restore failure, source shell
removal failure, nested-directory and simulated-symlink children, and
post-rename content drift — plus the retained-destination notice for a
dependent uncertain target in skill setup. The POSIX file identity impl
now consults the lstat seam so its degradation branches are coverable
the same way. Verified against the gate's own changed-line computation:
zero uncovered changed statements in internal/upgrade.
The publish-confirmation and tunneled-replacement tests physically
removed and reseeded the destination to simulate a concurrent swap. On
NTFS the recreation can immediately reuse the freed MFT record, making
the file ID (volume serial + file index) compare equal and the proof
pass against a replaced object — the Windows coverage gate observed the
confirmation falling through to the fingerprint branch instead of the
identity branch. Both tests now force the replacement through the two
primitives the platform proof consults (os.SameFile on Unix, the file-ID
seam on Windows), matching the technique the tunneled-rollback case
already used for Unix inode recycling.
skill_publication_identity_linux_test.go pinned the remote line's
Statx/birth-time identity design (skillPathStatx seam); the merged head
proves ownership with dev:ino plus the fingerprint backstop instead, so
the test no longer compiles on Linux. Caught by CI's Linux lint job,
which builds what macOS-local vet skips behind the linux build tag.
- Reject --request payloads with a missing, empty, or non-string
processCode; the backend answers a bad processCode with success:true
and an empty list, so validate client-side like startTime
- Add regression cases to keep changed-code coverage at 100%
Reconciles the two parallel evolutions of PR #996 with this session's
publication design as authoritative:
- internal/upgrade, internal/app: ours — mkdir-claim identity witness
(dev:ino on POSIX, volume file ID on Windows), three-state ownership,
ErrSkillPathPublicationUncertain, copy-fallback short-circuits. Drops
the remote line's xattr publication-mark design and its six follow-up
fixes (retract contracts, Statx token); skill_publication_mark_*.go
removed accordingly.
- scripts/, build/npm/, test/scripts/, docs/rfc: theirs — same replayed
install hardening plus main's evolution and the npm no-clobber child
moves; no xattr dependency, consistent with the claim model.
- .changes: their npm/Shell/PowerShell narrative with the Go-design
sentences rewritten for the uncertain-publication contract.
Verified: go build, go vet (tests compiled), gofmt, and package tests
for internal/upgrade, internal/app, test/scripts all green on this tree.
The mkdir->rename->remove->rename directory fallback had a TOCTOU window
between the second remove and the second rename: a concurrent writer
creating an entry at the destination was silently clobbered. The fallback
now claims the destination once with mkdir and never unlinks it: the
fast-path rename publishes over the claim (Linux), and platforms that
refuse directory renames (macOS, Windows) move the staged children into
the claim through atomic no-clobber primitives (mkdir/os.Link/os.Symlink),
consuming the emptied source shell on success.
renameSkillPathNoReplace now returns the mkdir-claim identity captured by
the child-move path. PublishSkillPathNoReplace uses it as a three-state
ownership witness: the atomic/fast paths keep the staged-inode proof, the
child-move path proves dest is still the mkdir claim, and a mismatch
reports the new ErrSkillPathPublicationUncertain sentinel with the
destination retained. The witness is real on POSIX now: darwin and linux
report the dev:ino file identity instead of the empty no-op.
Upstream consumers honor the sentinel: the mono/multi upgrade copy
fallbacks no longer retry over an uncertain destination (the retry would
displace the concurrent writer's object), and skill setup reports the
retained destination instead of claiming a rollback.
Rewrites the fallback tests that pinned the removed remove-and-retry flow
and adds regression coverage: concurrent claim entries abort with the
destination retained, wholesale replacement after child-move reports the
uncertain sentinel, staged-set transactions pass the sentinel through,
and both copy fallbacks short-circuit (ablation-verified).
- Reject --request payloads missing startTime (documented required) so
endTime can no longer bypass validation when startTime is absent
- Align --request time ordering with simple mode: endTime must be
strictly after startTime
- Cover all five previously uncovered branches (pageSize absent,
startTime absent, malformed endTime, valid time pair, empty --start
flag) to reach 100% changed-code coverage
- Add dws oa approval list-by-admin leaf with simple flags and
advanced --request modes backed by get_process_instances_by_admin
- Send startTime/endTime as yyyy-MM-dd HH:mm:ss strings per the
2026-08 MCP contract update; ISO-8601 flag inputs auto-convert
- Enforce pageSize cap (20) and string time format/order client-side;
PreRunE reports flag-group violations in Chinese before Cobra's
built-in English validation
- Extend coverage tests and document the command in mono/multi OA
skill references
Windows coverage gate only runs TestCrossPlatformCoverage*, and the
xattr mark helpers are Unix-only. Inject seams so marked dest is
retracted on owned drift, left in place when the mark is gone, and
the helper error paths are exercised on every platform.
Linux overlayfs recycles device+inode, so SameFile and a lone inode
token treated a replacement as owned and retracted it. Stamp staged
inodes with an xattr mark, prove Linux/Darwin identity with birth
time, and make shell copied-set rollback check dest first with inode
plus child names.
Match the Go dest-first identity check so a concurrent replacement is
never moved into .rollback-*; only a post-quarantine mismatch is
restored with no-replace. Cover both races in the npm smoke suite.
Record dest on occupy and retract it when confirmation, verify, or
staging cleanup fails. Restore unmatched quarantine with a no-replace
publish. Event/devapp copy uses mkdir-claim; shell rollback claims dest
before delete. Release copy now says npm/PowerShell create junctions and
Go uses os.Symlink, with copy fallback when linking is unavailable.
Cross-filesystem Skill moves now record publication identity as soon
as the staging path is renamed onto dest. A later mode-restore, copy
verification, or staging-cleanup failure retracts that proven dest so
retries are not blocked by an untracked leftover. A failed retract
reports an uncertain state naming both retained locations.
The shell mono/multi set publishers staged each Skill directory and
published it with a plain mv after the backup; anything another process
created at the destination between the backup and the move was silently
replaced, and restore_multi_skill_set then blind-deleted manifest paths,
so a concurrently replaced object could also be destroyed during
rollback. Publish through an atomic mkdir claim instead — EEXIST refuses
any occupant, staged children move into the claim one by one, and a
failed child move relocates them and removes only the claim. The
published manifest now records <dest>:<inode>, and rollback deletes a
destination only when its inode still matches the publication, skipping
concurrently replaced paths with a warning. Also fixes a latent
unbound-variable expansion where a shell variable was followed directly
by a full-width parenthesis in a message. Regression tests publish a
first Skill, replace it with a foreign directory, fail the second
publication, and assert rollback retains the foreign object untouched
while restoring the rest from backups.
The RFC section on filesystems that reject the atomic no-replace rename
still described the retired existence-check-plus-plain-rename fallback
and its accepted race window. The implementation (and the npm and shell
surfaces) claim the destination with mkdir or a hard link — or create
the link directly at the destination — and never release the claim mid
transaction, so a concurrently created object is refused rather than
overwritten. Record that contract and its only relaxed property (child
moves are not all-or-nothing visible) so future maintainers do not
port the racy description back into code.
- add success and failure outcomes for three attachment commands
- define business data schemas and mark downloadUri as sensitive
- migrate attachment commands to unified result output
- verify compact and full Schema result projections
- cover success, malformed response, and tool error paths
to #666
The POSIX shell installers staged shared Skill links and published them
with mv after an existence check; a file or symlink another process
created at the destination between the check and the move was silently
replaced, and the inode confirmation could not detect the loss. Publish
by creating each link directly at its destination instead — symlink(2)
refuses an occupied path with EEXIST, so the creation itself is the
atomic no-replace check. A directory that appears at the destination
turns ln -s into a container; the nested link is removed after an
identity check and the transaction rolls back, leaving the foreign
directory untouched. Applied to install.sh, install-skills.sh,
install-event.sh, and install-devapp.sh. Also covers the remaining
retraction branches of the Go shell-removal fallback so changed-code
coverage is complete. Regression tests inject a concurrent occupant at
the publish instant for regular-file and directory cases and assert the
foreign object and its contents stay completely unchanged.
On filesystems without atomic no-replace rename, the degraded
publication moves the source children into a fresh claim and leaves an
emptied source shell for the caller to remove once the move is
confirmed. If that removal failed, moveSkillPathRecoverably reported a
plain failure claiming both locations were preserved while the data
existed only at the destination, so backupAndRemoveSkillDir never
recorded the backup and the original path was left empty. Move the
children back into the shell and withdraw the destination instead; a
failed retraction reports the data location explicitly. Restores the
contract that a failed move keeps the source intact.
publishCanonicalLinkNoReplace checked the destination with lstat and
then published, leaving a window the comment claimed did not exist: on
Windows renameSync replaces a concurrent object outright (libuv passes
MOVEFILE_REPLACE_EXISTING), and on POSIX ln -P source target links INTO
a directory that appeared at the target, leaving a stray link inside
foreign data that the rollback list never recorded. Create the symlink
or junction directly at the destination instead — link creation fails
with EEXIST when anything occupies the path and never treats the target
as a container, so the publication itself is the atomic no-replace
check. Identity confirmation re-reads the live link before the
publication enters the rollback list. Covered by injected concurrent
creators at the publish instant on POSIX and simulated Windows,
asserting the foreign object and its contents stay completely
unchanged.
The mono and multi set copy publishers checked destination existence
with lstat and then called Node's rename, which replaces the target on
every platform (libuv passes MOVEFILE_REPLACE_EXISTING on Windows). A
file, symlink, or empty directory created between the check and the
rename was silently overwritten, and the identity confirmation could not
recover it because the publication record only proved the staged object
arrived. Claim the destination with mkdir — which fails with EEXIST if
anything occupies the path, so the claim itself is the existence check —
and move the staged children into the claim, restoring the source mode
on it. A failed child move relocates the children back and removes only
the claim. Covered for mono, multi, and simulated Windows, including an
injected concurrent creator at the claim instant.
The no-replace file fallback links the destination and then removes the
source. If the removal fails, the caller treats the publish as failed,
but no publication record exists to roll the new destination back, and
a backup restore would refuse the occupied path. Remove the destination
behind an identity check — only the proven linked object may be deleted
— and report when the retraction itself fails or the destination was
concurrently replaced.
The previous wrapper returned the first observed ID for both paths, so
expected == actual still held on Windows and the proof accepted the
swap. Return a distinct ID for the second probe.
The constant file-ID stub made both IDs equal, so the Windows proof
(expected == actual) accepted the publication and the subtest failed
there; Unix stayed green because its proof ignores the ID strings and
the swapped os.SameFile seam already forced the failure. Return the
first observed ID for both paths so staged and published identities
differ on every platform while real IDs still flow through the wrapper.
The physical same-content swap relied on the recreated destination
getting a fresh inode, but CI runners' ext4/overlayfs recycle inodes
eagerly, so the swap was undetectable on Linux and the subtest failed
there (while passing on macOS). Swap the same-file identity seam instead
so the confirmation's fast-path rejection contract is pinned on every
platform.
The four standalone installers pruned the oldest excess stamp
directories regardless of origin, so a migration retiring more than
five batches destroyed its own rollback material mid-run — the same
data loss already fixed for Go via the run-root registry and present
in install.js/install.ps1 as currentRunBackupRoots. Every installer now
records the stamp directories it creates and pruning only removes
earlier-run batches, which is what the changelog already promises.
The degraded directory publication claimed the destination with mkdir, then
— on platforms whose rename refuses to replace a directory (macOS refuses
even an empty target, verified empirically) — removed the claim and retried
a plain rename. Between the unlink and the retry a foreign directory could
appear at the destination and be silently overwritten, breaking the
no-replace contract the fallback exists to provide.
Hold the claim for the whole transaction instead: rename over the claim
where the platform permits it (Linux), otherwise move the source children
into the claim one by one. The destination is never unlinked, so a
concurrent creator can only ever lose the mkdir race; every child rename
targets a nonexistent path inside the empty claim, and a failed move
restores the children and removes only the claim.
The child move legitimately changes the publication's identity, which the
confirmation now handles: a rename that consumed the staged path is still
proven by identity, while a child move is proven by the pre-rename content
fingerprint. The emptied source shell doubles as the signal distinguishing
the two shapes; moveSkillPathRecoverably removes it to keep move semantics.
The backup stamp has second precision and pruning kept only the newest 5
stamps, so a canonical migration that retires copies across many Agent
roots deleted its own earlier backups mid-run. That silently voided the
reversibility guarantee the transaction depends on for rollback: a probe
retiring 8 paths lost 3 of them permanently.
Record every stamp directory this process creates, keyed by normalized
absolute path, and prune only the oldest foreign stamps.
Creating the staged symlink usually succeeds, so the link strategy really
fails at publish time: renameSkillPathNoReplace has no atomic no-clobber
primitive for a symlink source and refuses it whenever the kernel flag is
unavailable (NFS, FUSE, overlayfs). Gating the copy fallback on staging
alone therefore left every non-universal Agent unconfigured on exactly the
filesystems the fallback exists to support.
Retry the whole target transaction as a direct copy after a failure in any
phase, but only when the failed attempt fully restored the originals. The
converter also re-adds the replacement backups the link plan deliberately
skips for destinations already pointing at canonical, which a copy must
replace and no-replace publication would otherwise reject with EEXIST.
skillPathSameFileIdentityImpl on Windows always returns false and is
never reached through skillPathIdentityProven (which uses file IDs
exclusively). Add a direct seam call with synthetic os.FileInfo to
exercise the Windows return-false path and the Unix os.SameFile path
with nil Sys().
Restructure skillPathFileIdentityImpl to use nested if-err-nil with a
named return and skillPathIdentityProven to use a single expression.
Error conditions now fall through to the bare return instead of
occupying separate coverage blocks, eliminating 5 uncovered statements
that the Windows coverage gate flagged at 99.4193%.
Add FILE_FLAG_OPEN_REPARSE_POINT to the Windows CreateFile call in
skillPathFileIdentityImpl so symlinks are opened as reparse points
rather than followed to their target. Staged symlinks carry relative
targets computed for the final destination, which may not resolve from
the staging directory; following them caused CreateFile to fail,
yielding an empty file ID that rejected publication and broke canonical
skill layout migration on Windows.
Make skillPathSameFileIdentity a seam variable so the tunneled
replacement test can deterministically simulate the identity change on
Unix. On tmpfs (used by Linux CI runners), os.SameFile can return true
for a recreated file due to inode reuse, making the test flaky. On
Windows the swap is a no-op because skillPathIdentityProven compares
file IDs from GetFileInformationByHandle and ignores
skillPathSameFileIdentity.
NTFS file tunneling can restore the original creation time for a
recreated same-named object, which defeated the creation-time
incarnation check and allowed rollback to delete a concurrent
replacement. Replace the platform-specific identity pair with a single
skillPathIdentityProven function:
- Unix: delegates to os.SameFile (inode/dev), ignoring file ID strings
- Windows: compares VolumeSerialNumber:FileIndexHigh:FileIndexLow from
GetFileInformationByHandle, which uniquely identifies the file on the
volume for its lifetime and is unaffected by tunneling
When the file ID cannot be obtained at publish time, identity is not
proven and the auto-delete is refused. Add a regression test that
simulates tunneled creation time and verifies rollback still refuses
the concurrent replacement.
On Windows isNoReplaceRenameUnsupported always returns false, so the
fallback is never entered from the invalid-path test. Force the fallback
and swap skillPathLink to a non-EEXIST error to cover line 94 on all
platforms.
Add tests for mkdir non-EEXIST error, remove failure after rename
failure, first-rename-succeeds path (Linux behavior), retry-rename
path, and non-regular source safe-fail. All 24 changed executable
statements now covered on both macOS and Windows.
The fallback path for filesystems without RENAME_NOREPLACE/EXCL (NFS,
FUSE, overlayfs) used Lstat-then-Rename, which could overwrite a
concurrently created destination between the check and the rename.
Replace the TOCTOU-prone check with truly atomic no-clobber primitives:
- Directories: os.Mkdir atomically claims the destination (fails with
EEXIST if occupied). On Linux rename(2) replaces the empty dir
directly; on Darwin/Windows rename refuses existing dirs so the empty
dir is removed and the rename retried — any concurrent creation
between remove and rename is detected by the second rename failing.
- Files: os.Link atomically fails if the destination exists, then
os.Remove completes the move.
Add concurrent-creation test covering the mkdir→rename race window.
The !os.IsNotExist(statErr) branch in renameSkillPathNoReplace was
uncovered on Windows. Inject errNoReplaceRenameUnsupported for the
atomic rename and os.ErrPermission for skillPathLstat so the stat-error
path is exercised on every platform.
The smoke test creates temp home directories with .config/kimchi markers
for agent detection. On Linux CI runners XDG_CONFIG_HOME may point to the
runner's real config path, causing resolvedAgentTargets to look outside
the temp home. Unset it so detection resolves against the test's temp dir.
Restrict backup pruning to directories whose names match the DWS stamp
format (YYYYmmdd-HHMMSS with optional -N suffix) across all 8 installer
surfaces (Go, npm, 4 shell, 2 PowerShell). Unknown directories in
~/.dws/skill-backups are now preserved. Also fixes Windows coverage test
portability and covers the remaining macOS changed-code gap (retire
warning loop in runUpgrade).
A universal Agent whose obsolete private copy cannot be retired installs
nothing there, yet every entry point counted that retirement failure as an
install failure — aborting `npm install`, `dws skill setup`, and the shell
installers even when the canonical store and all links published correctly,
and skipping the skills-state write. Route retirement failures to a separate
warning path across all surfaces (Go upgrade + skill setup, npm, PowerShell,
install.sh, install-skills.sh, install-event.sh, install-devapp.sh).
Also:
- Add a checked-rename fallback for filesystems that reject the atomic
no-replace flag (NFS, FUSE, overlayfs); the no-clobber contract is kept and
the previously unsupported platforms build and work.
- PowerShell multi-mode links only bundle skills, never the shared canonical
store, so third-party/user skills are no longer fanned into every Agent root.
- Prune ~/.dws/skill-backups to the newest 5 on every surface; encode
HOME-relative backup names on PowerShell to preserve origin.
- Add simulated-win32 junction coverage and rewrite the tautological
no-replace test; remove dead code whose tests gave false coverage.
- Soften the overstated Windows ownership-proof comment (NTFS tunneling).
- extend the silent-rollback contract to install-event.sh
- static contract: Restore-MultiSkillSet removes published paths lexically
(section-scoped so identity-anchor refactors keep the guarantee) and link
staging dirs are cleaned via Remove-LinkStageRoot / Remove-DevLinkStageRoot
- install-event.sh integration test: an uninstallable agent target is
skipped loudly while later agents still receive links
- pwsh probe: Test-SamePhysicalSkillRoot must dereference junctions and
symlinks (junction idempotency asserted where junctions are creatable)
- install.ps1: remove published junctions lexically in Restore-MultiSkillSet
(Windows PowerShell 5.1 follows reparse points during Remove-Item -Recurse
and could delete canonical store contents); clean link staging dirs
lexically in Publish-CanonicalSkillLinks and Move-SkillPathRecoverably
- install.ps1: Test-SamePhysicalSkillRoot now dereferences junctions via
Get-PhysicalSkillPath (mirrors EvalSymlinks/realpathSync/cd -P), so reruns
recognize already-published junctions instead of backup churn
- install-event.sh: replace silent 'mv ... 2>/dev/null || true' rollback with
the loud backup-retained failure contract already enforced for devapp
- event/devapp sh+ps1: link→copy fallback and per-agent failures now degrade
per agent like install.sh (skip loudly, continue, report at the end)
instead of aborting mid-loop or swallowing errors
- tests: junction-lexical removal contract, event per-agent degrade
integration test, pwsh junction physical-root recognition + rerun
idempotency (no backup churn)
A failed canonical publish only failed the upgrade when
hasDependentSkillRoot reported a non-universal link target; that helper
explicitly skipped universal agents, which are exactly the direct consumers
of ~/.agents/skills. On a universal-only machine (e.g. only Codex
installed), UpgradeSkillLocations* returned a nil error with nothing
installed, contradicting the documented "canonical publication is
mandatory and fails the upgrade loudly" contract.
Canonical publish failures now return an error unconditionally in both the
mono and multi branches, and hasDependentSkillRoot is removed. The test
that pinned the old standalone-does-not-fail-fast behavior now asserts
error propagation in both modes.
The allowSystemApps gate (homeDir == systemHome) was effectively a no-op in
production: systemHome came from os.UserHomeDir, which honors the $HOME env
override just like homeDir, so the two were always equal and the gate never
fired when $HOME was overridden.
ResolveSystemHomeDir now prefers the OS user database (getpwuid on Unix),
which is independent of $HOME, falling back to $HOME only when the user record
cannot be resolved. Production behavior is unchanged (a real $HOME still
matches); an isolated/overridden HOME now correctly skips machine-wide
/Applications discovery for zcode/minimax. The app surface references the same
shared resolver.
This is the correct fix for the hermeticity concern (machine-wide state leaking
into an isolated HOME): there is no cross-surface production inconsistency to
port — script installers always operate on the real user HOME in practice, so
they need no gate.
skillPathSameFileIdentityImpl on Windows always returns false and is
never reached through skillPathIdentityProven (which uses file IDs
exclusively). Add a direct seam call with synthetic os.FileInfo to
exercise the Windows return-false path and the Unix os.SameFile path
with nil Sys().
Restructure skillPathFileIdentityImpl to use nested if-err-nil with a
named return and skillPathIdentityProven to use a single expression.
Error conditions now fall through to the bare return instead of
occupying separate coverage blocks, eliminating 5 uncovered statements
that the Windows coverage gate flagged at 99.4193%.
Add FILE_FLAG_OPEN_REPARSE_POINT to the Windows CreateFile call in
skillPathFileIdentityImpl so symlinks are opened as reparse points
rather than followed to their target. Staged symlinks carry relative
targets computed for the final destination, which may not resolve from
the staging directory; following them caused CreateFile to fail,
yielding an empty file ID that rejected publication and broke canonical
skill layout migration on Windows.
Make skillPathSameFileIdentity a seam variable so the tunneled
replacement test can deterministically simulate the identity change on
Unix. On tmpfs (used by Linux CI runners), os.SameFile can return true
for a recreated file due to inode reuse, making the test flaky. On
Windows the swap is a no-op because skillPathIdentityProven compares
file IDs from GetFileInformationByHandle and ignores
skillPathSameFileIdentity.
NTFS file tunneling can restore the original creation time for a
recreated same-named object, which defeated the creation-time
incarnation check and allowed rollback to delete a concurrent
replacement. Replace the platform-specific identity pair with a single
skillPathIdentityProven function:
- Unix: delegates to os.SameFile (inode/dev), ignoring file ID strings
- Windows: compares VolumeSerialNumber:FileIndexHigh:FileIndexLow from
GetFileInformationByHandle, which uniquely identifies the file on the
volume for its lifetime and is unaffected by tunneling
When the file ID cannot be obtained at publish time, identity is not
proven and the auto-delete is refused. Add a regression test that
simulates tunneled creation time and verifies rollback still refuses
the concurrent replacement.
On Windows isNoReplaceRenameUnsupported always returns false, so the
fallback is never entered from the invalid-path test. Force the fallback
and swap skillPathLink to a non-EEXIST error to cover line 94 on all
platforms.
Add tests for mkdir non-EEXIST error, remove failure after rename
failure, first-rename-succeeds path (Linux behavior), retry-rename
path, and non-regular source safe-fail. All 24 changed executable
statements now covered on both macOS and Windows.
The fallback path for filesystems without RENAME_NOREPLACE/EXCL (NFS,
FUSE, overlayfs) used Lstat-then-Rename, which could overwrite a
concurrently created destination between the check and the rename.
Replace the TOCTOU-prone check with truly atomic no-clobber primitives:
- Directories: os.Mkdir atomically claims the destination (fails with
EEXIST if occupied). On Linux rename(2) replaces the empty dir
directly; on Darwin/Windows rename refuses existing dirs so the empty
dir is removed and the rename retried — any concurrent creation
between remove and rename is detected by the second rename failing.
- Files: os.Link atomically fails if the destination exists, then
os.Remove completes the move.
Add concurrent-creation test covering the mkdir→rename race window.
The !os.IsNotExist(statErr) branch in renameSkillPathNoReplace was
uncovered on Windows. Inject errNoReplaceRenameUnsupported for the
atomic rename and os.ErrPermission for skillPathLstat so the stat-error
path is exercised on every platform.
The smoke test creates temp home directories with .config/kimchi markers
for agent detection. On Linux CI runners XDG_CONFIG_HOME may point to the
runner's real config path, causing resolvedAgentTargets to look outside
the temp home. Unset it so detection resolves against the test's temp dir.
Restrict backup pruning to directories whose names match the DWS stamp
format (YYYYmmdd-HHMMSS with optional -N suffix) across all 8 installer
surfaces (Go, npm, 4 shell, 2 PowerShell). Unknown directories in
~/.dws/skill-backups are now preserved. Also fixes Windows coverage test
portability and covers the remaining macOS changed-code gap (retire
warning loop in runUpgrade).
typedJSONValue marshaled a typed value and then routed the result through
rawJSONValue, which runs json.Valid before decoding. On that path the input is
whatever json.Marshal has just produced, so the validation scan can only ever
succeed: it re-read every marshaled document for nothing.
The decode step is now shared by both entry points. rawJSONValue keeps its
json.Valid check, because it still accepts untrusted input, while typedJSONValue
decodes what it marshaled directly. Across the 1121-tool set this removes about a
third of the Schema Catalog projection work: the internal/app schema suite goes
from 26.0s to 17.2s uninstrumented, and from 291.1s to 241.0s under -race.
The delivered Catalog is byte-for-byte unchanged. check-generated-drift,
check-schema-catalog and check-schema-binary each regenerate the same
source_hash sha256:93b8d44eb163bd2898c78397d22af92d378e3dc4e20f56b33277b51e4342e2e6,
and the two error contracts are preserved: typedJSONValue still rejects a value
json.Marshal cannot encode, and rawJSONValue still rejects invalid JSON.
The five internal/app partitions ran end to end inside one job, so the app
shard's wall clock was the sum of all five: 780s in CI, of which the schema
partition owned 357s. Each partition is now its own matrix shard, so they run
concurrently and the shard's wall clock is set by its slowest partition rather
than by their total. Every partition shard still selects the same single
internal/app package, so the impacted-package query maps the shard name back to
app and the partition only chooses which tests run.
The helper gains a partition argument and a list-partitions mode. APP_PARTITIONS
is the single source of truth for the set, and the discovery pass still runs in
every job, so each one independently verifies that the partition patterns cover
every top-level test exactly once before running the one it was asked for.
Two fail-closed checks guard the split, because the helper's own coverage check
can no longer prove the whole package ran once the partitions are separate jobs:
- The helper cross-checks APP_PARTITIONS against the coverage counters in both
directions, so a counted partition that nothing dispatches and a dispatchable
partition with no counter both fail instead of silently skipping tests.
- TestCIAppRacePartitionMatrixMatchesHelper pins the workflow's app-<partition>
shards to list-partitions output in both directions, so a partition cannot
lose its job while every job stays green.
The discovery loop variable is renamed from partition to spec: it would
otherwise shadow the partition requested on the command line, which run mode
reads after the discovery pass completes.
The schema partition's 52 tests assert structural Schema-to-Cobra contracts over
a single goroutine: none of them call t.Parallel or start a goroutine, so the
race detector has no concurrent access to observe there. The process-global lazy
metadata that does need race coverage (schema_source_root's atomic.Value, the
parameter-binding lazy loaders) is exercised by internal/cli's concurrent tests,
which stay instrumented.
The instrumentation was not free here. The partition shares a single sync.Once
Catalog build whose work is allocation-heavy, and -race made it roughly 11x
slower: 26s -> 291s locally, and 357s of the app shard's 780s in CI. Within that
partition TestFinalSchemaToolsHaveExecutableBaseCommands alone accounted for
262s, not because the test is expensive but because it is the first caller to pay
for the shared snapshot; its 1121 subtests together measure 0.00s.
run_partition now takes the instrumentation mode explicitly and fails closed on
an unrecognized value, so a typo cannot silently drop -race from a partition that
is supposed to carry it.
The workflow contract pinned the focused path by literal: the job name
`Test (changed packages)`, the unsharded
`list "$TEST_BASE_REF" "$TEST_HEAD_REF"` call, and a single
`go test -timeout=15m` line standing in for internal/app's package-level
headroom. Sharding the job changed all three literals, so `Test (workflow
and release contracts)` failed on this branch even though every shard
selection test passed.
Each invariant the contract guarded still holds, so the assertions are
updated to the new shape rather than relaxed:
- the focused job must still exist, now as the matrix job, named the way
the contract already names `Test (race: ${{ matrix.shard }})`;
- package selection must still derive from the authoritative synthetic
merge base/head, now with an explicit shard argument, so pointing it at
any other ref still fails the contract;
- internal/app's headroom is asserted through the process-isolating
helper and the per-shard budgets, mirroring the assertions already
applied to test-race. That is stronger than the old single -timeout: it
pins the mechanism that keeps the suite inside its budget rather than
the number alone. release-scripts membership is asserted too, because
its dedicated job only runs at full-suite or release-sensitive scope,
so losing it here would silently stop testing test/scripts changes.
The shard comparisons in the focused job are quoted so that job reads
verbatim like test-race's.
Ablating the implementation one change at a time turns the contract red
in all five cases: removing the app helper call, dropping release-scripts
from the matrix, selecting from HEAD~1, collapsing the matrix back to a
single unsharded job, and dropping the cli/smoke timeout budget.
Reading the package list with `mapfile < file` has unambiguous line
semantics. Routing it through a step output and a here-string instead
would append an extra empty array element if the value ever carried a
trailing newline, and that element would reach go test as an empty
package argument. The step output now carries only a single-line boolean,
and the list travels through RUNNER_TEMP. An explicit empty-entry guard
fails closed if the file is ever malformed.
This job cannot execute on its own pull request — editing a workflow
routes the revision to full_suite, which skips the focused path — so the
implementation deliberately avoids depending on platform-specific
trailing-newline behavior that local verification cannot observe.
The focused path tested every impacted package in a single job with a
plain `go test -race`, so internal/app ran inside one long-lived process
alongside all of its reverse dependencies. That is exactly the shape
scripts/ci/run-app-race-tests.sh exists to avoid: a single app test
process retains every constructed command tree in framework registries,
so the run grows to 900s and the job stays alive long enough to be
reclaimed by the runner. Recent focused runs failed with SIGTERM after
9-10 minutes without a single test failure, and one earlier run failed
at `internal/app 902.651s`, 2.65s past the package timeout.
Fan the same package plan across the shard matrix test-race already
uses, and run each shard the way test-race runs it: internal/app through
the process-isolating helper, cli/smoke with their wider package budget,
release-scripts without race and with archive tooling.
changed-test-packages.sh gains `list-shard`, which intersects the
impacted set with scripts/ci/test-packages.sh shard membership so shard
definitions stay single-sourced — and so an unknown shard name aborts
there rather than reporting an empty selection, which would let a
mistyped shard skip every test while reporting success.
release-scripts is in the matrix on purpose: its dedicated job only runs
at full-suite or release-sensitive scope, so omitting it here would stop
testing test/scripts changes altogether. A test pins that the shard
selections partition the impacted set exactly, so shard-plan drift
cannot silently shrink focused coverage.
Every neighbouring string in the root help listing — service
descriptions, utility descriptions, global flag usage — is hardcoded
Chinese. Routing only the feedback label through i18n therefore rendered
it in English on any host whose LANG is not zh_*, leaving a lone English
line inside an otherwise Chinese screen.
Hardcode the label and drop the two locale entries it needed. A test
assertion now pins the Chinese label so the indirection cannot return
unnoticed.
`dws --help` now closes with a Feedback section that links the
user-experience survey form, tagged with source=dws-cli so submissions
arriving through the CLI can be told apart from other channels.
The entry is deliberately root-only: this CLI is driven mostly by AI
agents, and repeating a survey link in every subcommand help would be
pure context noise. A guard test pins that boundary.
The URL is printed on its own unwrapped line — it is longer than the
help rule width, and breaking it would stop terminals from recognizing
it as a clickable hyperlink.
A universal Agent whose obsolete private copy cannot be retired installs
nothing there, yet every entry point counted that retirement failure as an
install failure — aborting `npm install`, `dws skill setup`, and the shell
installers even when the canonical store and all links published correctly,
and skipping the skills-state write. Route retirement failures to a separate
warning path across all surfaces (Go upgrade + skill setup, npm, PowerShell,
install.sh, install-skills.sh, install-event.sh, install-devapp.sh).
Also:
- Add a checked-rename fallback for filesystems that reject the atomic
no-replace flag (NFS, FUSE, overlayfs); the no-clobber contract is kept and
the previously unsupported platforms build and work.
- PowerShell multi-mode links only bundle skills, never the shared canonical
store, so third-party/user skills are no longer fanned into every Agent root.
- Prune ~/.dws/skill-backups to the newest 5 on every surface; encode
HOME-relative backup names on PowerShell to preserve origin.
- Add simulated-win32 junction coverage and rewrite the tautological
no-replace test; remove dead code whose tests gave false coverage.
- Soften the overstated Windows ownership-proof comment (NTFS tunneling).
Fifth-review addition: declaring InputFile silently claims the whole
@-prefixed value space, which matters in this product because at-mention
style values are common (--at-user @zhangsan would report a file read
failure), and declaring InputStdin makes a literal "-" unreachable. Both
are decided at declaration time and cannot be fixed downstream, so record
them next to the confirmation rule in the author rules.
Fourth-review fix: the FlagSpec sub-field table in the homology doc is
the named authority for "what each field does and whether it reaches
Schema parameters", and RFC §5.0.2 asserts declaration fields embed into
dws.schema.*. Input satisfied neither entry, leaving its deliberate
non-projection indistinguishable from an oversight. Add the table row and
the §5.0.2 exception note so the capability stays a declared fact (Usage
prose) rather than inviting an invented annotation.
Third-review fix for a CI blocker: run-platform-coverage-gate.sh only
executes ^(TestAllShortcuts|TestCrossPlatformCoverage) yet enforces 100%
coverage of changed production lines, so the TestResolveInputFlags names
left every new input.go statement reported as uncovered. Rename them to
the gate prefix, drop three unreachable pflag Set error branches that no
test could ever cover, and add the reachable stdin read-failure case.
Verified: changed code coverage 100.0000% (67 statements).
Second-review fix: explicitInputFlagName judged usability with an
unconditional TrimSpace while rawValue only trims when Trim is set. For
a non-Trim flag a whitespace main value is usable and shadows a changed
alias; the resolver could then rewrite the shadowed alias (and fail on
its @path) while the fallback chain still read the main value. Mirror
rawValue's usable() exactly and pin the shadow case with a regression
test whose alias path does not exist.
Self-review fixes: a Trim flag receiving " @path" judged usability on the
trimmed value (rawValue) while the source prefix check saw the raw value,
so the token would ship as a literal. Trim before the prefix check. Also
build the file-read error once with a conditional hint option, and pin
the default-value/env passthrough plus Trim edge with regression tests.
Document the landed corecmd.Input transitional form: declaration shape
(FlagSpec/LeafFlag/shortcut.Flag), runtime resolution semantics and
ordering, author rules (help prose, confirmation interaction with
stdin, construction-time validation), and the delta table against the
target typed InputSource design.
Port the lark-cli Flag.Input capability: a KindString flag may declare
Input sources ("file" for @path, "stdin" for -) and the framework
rewrites the explicit token into the payload content before
required/enum/constraint/Validate checks. @@value escapes to a literal
@value; a single stdin consumer per invocation is enforced; a leading
UTF-8 BOM is stripped. Shortcut.Flag gains the same declaration and the
adapter maps it through; LeafSpec inherits it via the LeafFlag alias.
Split chat message and group references by task, update intent routing and context budget, distinguish accepted card updates from verified writes, and explain the ambiguous chat --from flag.
- extend the silent-rollback contract to install-event.sh
- static contract: Restore-MultiSkillSet removes published paths lexically
(section-scoped so identity-anchor refactors keep the guarantee) and link
staging dirs are cleaned via Remove-LinkStageRoot / Remove-DevLinkStageRoot
- install-event.sh integration test: an uninstallable agent target is
skipped loudly while later agents still receive links
- pwsh probe: Test-SamePhysicalSkillRoot must dereference junctions and
symlinks (junction idempotency asserted where junctions are creatable)
- install.ps1: remove published junctions lexically in Restore-MultiSkillSet
(Windows PowerShell 5.1 follows reparse points during Remove-Item -Recurse
and could delete canonical store contents); clean link staging dirs
lexically in Publish-CanonicalSkillLinks and Move-SkillPathRecoverably
- install.ps1: Test-SamePhysicalSkillRoot now dereferences junctions via
Get-PhysicalSkillPath (mirrors EvalSymlinks/realpathSync/cd -P), so reruns
recognize already-published junctions instead of backup churn
- install-event.sh: replace silent 'mv ... 2>/dev/null || true' rollback with
the loud backup-retained failure contract already enforced for devapp
- event/devapp sh+ps1: link→copy fallback and per-agent failures now degrade
per agent like install.sh (skip loudly, continue, report at the end)
instead of aborting mid-loop or swallowing errors
- tests: junction-lexical removal contract, event per-agent degrade
integration test, pwsh junction physical-root recognition + rerun
idempotency (no backup churn)
The Coverage context was the PR critical path (~17 min end to end):
coverage-current re-ran the whole suite serially (-p 1, ~13 min) and
coverage-baseline re-ran it again at the merge-base (~13 min) although
that profile is a pure function of the base commit.
- coverage-current now owns only the scoped (standard-tier) profile;
full-suite candidate profiles come from a 5-way shard matrix
(app/cli/generators/helpers/remaining) that keeps -p 1 inside each
shard on isolated runners. scripts/ci/test-packages.sh list-coverage
defines the shards and verify proves the union equals the previous
single-run package set exactly once.
- the aggregate Coverage job reassembles the disjoint shard profiles
into coverage.txt before make coverage-gate, failing closed when a
shard file is missing, so gate semantics (100% changed-code +
scope-matched overall non-regression) are byte-compatible.
- coverage-baseline restores the merge-base full-suite profile from an
exact-key cache (merge-base SHA + resolved Go version) written by the
last green main push; any miss falls back to recomputing in the
merge-base worktree. Exact key only - no prefix fallback, a near-miss
profile would compare the candidate against the wrong commit.
- new contract tests pin the shard matrix, the assembly step, the
exact-key cache pair, and the absence of restore-keys; the package
plan test also covers the coverage shard partition.
- Fix Example indentation (tab -> 2 spaces)
- Remove unsubstantiated default en-US from --language help/docs
- Add test asserting calendarId/language are omitted when only --id is passed
A failed canonical publish only failed the upgrade when
hasDependentSkillRoot reported a non-universal link target; that helper
explicitly skipped universal agents, which are exactly the direct consumers
of ~/.agents/skills. On a universal-only machine (e.g. only Codex
installed), UpgradeSkillLocations* returned a nil error with nothing
installed, contradicting the documented "canonical publication is
mandatory and fails the upgrade loudly" contract.
Canonical publish failures now return an error unconditionally in both the
mono and multi branches, and hasDependentSkillRoot is removed. The test
that pinned the old standalone-does-not-fail-fast behavior now asserts
error propagation in both modes.
The allowSystemApps gate (homeDir == systemHome) was effectively a no-op in
production: systemHome came from os.UserHomeDir, which honors the $HOME env
override just like homeDir, so the two were always equal and the gate never
fired when $HOME was overridden.
ResolveSystemHomeDir now prefers the OS user database (getpwuid on Unix),
which is independent of $HOME, falling back to $HOME only when the user record
cannot be resolved. Production behavior is unchanged (a real $HOME still
matches); an isolated/overridden HOME now correctly skips machine-wide
/Applications discovery for zcode/minimax. The app surface references the same
shared resolver.
This is the correct fix for the hermeticity concern (machine-wide state leaking
into an isolated HOME): there is no cross-surface production inconsistency to
port — script installers always operate on the real user HOME in practice, so
they need no gate.
The new root-type guard shifted `filepath.WalkDir`'s outer error branch into
the diff, and neither the macOS nor the Windows runner reaches it naturally —
raising Windows coverage to 99.9365% and blocking the gate. Add a
`statusWalkDir` seam and a `TestCrossPlatformCoverage` regression that swaps
in a WalkDir returning a sentinel error, asserting it is surfaced unchanged.
Verified locally: changed code coverage back to 100.0000%.
Two follow-ups to the latest CR:
* push/sync uploads (`pushUploadFilePinned`): the PUT-time check pinned inode,
size, and mtime before dispatch but nothing rechecked the source after PUT
succeeded — only the root itself. An editor overwrite, truncate-rewrite, or
mmap-in-place during transfer would land a mixed old/new byte stream in OSS
and still be committed, corrupting the remote file in overwrite/local-wins.
Now stat the still-open handle again before `commit_upload`; any change in
inode/size/mtime aborts the commit. Post-PUT stat failures also abort.
* status root (`walkLocalTree`): `filepath.WalkDir` refuses to follow the root
when it is itself a directory symlink and reports it as a non-regular entry,
so the walker silently returned an empty local index and status flagged
every remote file as `new_remote`. Fail closed before the walk: the root
must be a real directory; symlinks and non-directories are rejected with a
clear message. A `statusRootLstat` seam keeps the rejection regressible on
platforms that cannot create directory symlinks (Windows without admin).
Both fixes come with `TestCrossPlatformCoverage*` regressions and take the
platform coverage gate from 99.9356% back to 100.0000% (1553 statements).
The Windows coverage gate reported changed-code coverage at 99.9360% because
drive_push.go:471-473 — the branch that surfaces an error passed to the
fs.WalkDir callback as its third argument — was not exercised. macOS runners
happen to exercise it via directory-lstat failures, Windows runners do not.
Add walk_callback_receives_error under
TestCrossPlatformCoverageDrivePushFinalWalkAndCommandGates, which swaps
walkPinnedLocalFS to invoke the callback with a non-nil err and asserts the
error is bubbled up unchanged.
Verified locally that the new subtest hits drive_push.go:471.17,473.4 with
count=1.
Windows keeps the pinned directory locked while a handle inside it is open
(os.Root plus the pull temp file or the upload source), so renaming that
directory fails with a sharing violation. Every "pinned root/ancestor was
swapped" reproduction in the drive mirror tests relied on such a rename, so 13
tests failed on windows-latest. That, not a coverage shortfall, is why
Coverage (Windows) exited 1 before the gate ever ran.
Each reproduction now falls back to injecting the equivalent identity change
when the rename is refused. pinnedPullRoot.verify() and verifyParent() read
current identity only through pullPathStat / pullRootLstat, so pointing those
seams at another directory hits the same fail-closed branches. Unix still
performs the real move and loses no strength.
Assertions that need an actual replacement tree now branch on the helper's
return value. forcePinnedFallbackForTest makes the fallback path itself
regressible on any platform, and a dedicated test covers it.
Verified locally with the fallback forced on: all 13 tests pass and changed
code coverage stays at 100%.
The platform coverage gate runs only TestAllShortcuts and
TestCrossPlatformCoverage*, so several changed statements had no platform
test exercising them:
- drive_pull.go: the smart-policy re-check that skips publication when the
target is refreshed in place (same inode) while the download is running.
- drive_pull.go: the post-publish verifyParent failure, where the result is
already on disk and must not be rolled back.
- drive_replace_unix.go: rename(2) replacement of an existing target; the
Windows side already had the symmetric test.
- drive_status_windows.go: the filepath.Clean rewrite guard had no input
reaching it, because isSafeRemoteSegment filters separators upstream.
macOS changed-code coverage: 99.8053% -> 100.0000% (1541 statements).
release_version was interpolated into an awk regex, where '.' matches any
character. Version 1.0.1-beta.1 therefore also admitted
.changes/released/1x0x1-betaX1/, letting the archive drift from the
CHANGELOG version while every other seal assertion still passed and
breaking the documented audit trail.
Compare the archive prefix with index() and split the basename off with
substr(), matching the literal-comparison idiom already used throughout
check-changelog-pr.sh. Only the basename, whose character class is fixed,
stays a pattern.
Both trigger predicates ran the same git diff, which the script already
avoids elsewhere by staging --name-status into $tmp_root/status. Write the
path list once and let each awk predicate read it, matching that idiom.
Git records no diff entry for a directory itself, so adding
.changes/foo/bar.md only surfaced the nested path, which the single-level
trigger regex skipped. The entry validation and the renderer were both
bypassed, letting a nested directory reach main and break every later
fragment render with 'unexpected directory'.
Trigger the top-level tree validation on any .changes change outside
.changes/released/ (which keeps its own immutability and release-seal
checks), and assert .changes itself is still a tree so replacing it with a
blob or symlink cannot empty the child listing unnoticed.
Re-rendering stays keyed on fragment changes so a README-only edit does
not fail on an empty fragment set.
Bind each accepted marker to the exact workflow run attempt, immutable artifact, source comment, and current PR head so a historical successful run cannot authorize a different payload.
The fragment gate only ran validation when the changed path matched the
legal fragment name pattern, so `.changes/Foo.md`, `.changes/notes.txt`
and a symlinked fragment slipped through untouched and then broke the
next PR that added a legal fragment. The trigger now fires on any
top-level `.changes/` change other than README.md and rejects every
entry that is not README.md, released/, or a 100644 blob named
^[a-z0-9][a-z0-9._-]*\.md$.
The renderer had the same hole from the other side: `find -type f`
is false for symlinks, so a symlinked fragment was silently dropped
from the rendered notes, and the `[a-z0-9]*.md` glob only constrained
the first character so `chat reply.md` passed. It now walks every
top-level entry and fails on symlinks, unexpected directories,
non-regular files and illegal names. Both scripts pin LC_ALL=C so the
ASCII ranges cannot match uppercase under a different collation.
Adds regression coverage for illegal names, non-markdown entries,
symlinks and executable modes on both the gate and the renderer.
Address P1 finding: extract_payload now strictly requires the parsed JSON
to be a dict, and validates each field's type and format:
- pr_number: string of digits
- pr_head_sha: 40-char lowercase hex string
- products: alphanumeric with commas/dots/hyphens/underscores only
- run_id: string of digits
- cases_ref: string (may be empty)
validate_run_id also guards against non-string input.
Added tests for: integer/array/string/null JSON, numeric field types,
invalid SHA format, injection in products, missing required fields.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Address P1 lint finding: structured eval-dispatch comments could be
forged by unauthorized users. Add three-layer consumer-side validation:
1. comment.user.login == 'github-actions[bot]' (platform-enforced identity)
2. comment.performed_via_github_app.slug == 'github-actions' (App signature)
3. payload.run_id verified against actual successful workflow run via API
Also adds:
- eval_poll_validate.py: consumer validation module (in-repo, auditable)
- test_eval_poll_validate.py: unit tests proving forged comments are rejected
- Go security contract test updated to assert run_id and validate reference
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The GitHub Actions runner cannot reach internal Aone CI API (structural
network isolation). Replace the curl-to-internal step with a structured
HTML comment (<!-- eval-dispatch: {...} -->) that an internal Devix
polling service picks up every 3 minutes to trigger the Aone CI pipeline.
This eliminates the EVAL_TRIGGER_URL/EVAL_TRIGGER_TOKEN secrets dependency
from the GitHub side — those can be removed once verified.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- /eval on one's own PR may omit sha=: the guard auto-pins the
dispatch-time head (commenter == PR author leaves no third-party
swap window); dispatching another author's PR still requires the
explicit reviewed SHA (keeps the P1-2 TOCTOU remedy where the
threat lives)
- cases= is now validated structurally per git check-ref-format
semantics (leading/trailing//double slashes, '..', dot-leading
components, .lock suffixes) and rejects '-'-leading values to
prevent git fetch option injection (review P2)
Users listed in .github/eval-allowlist.txt (default branch, PR-reviewed)
may dispatch /eval for their own PRs only; write/maintain/admin retain
dispatch for any PR. Fail-closed on permission API 404/network errors.
The two Minutes notes (permission apply --policy int typing and the skill
reference updates) landed in the released 1.0.58-beta.2 section after the
branch merged main. That rewrites published release notes and would drop
both notes from the next release generated out of Unreleased. Move them
verbatim into a Changed subsection under Unreleased; the beta.2 section is
byte-identical to main again.
Auto-CR (P1) correctly flagged that the carve-out's "nothing else changed"
guard was keyed on len(otherFailures) == 0, which only observes changes the
gate already judges incompatible. Several parameter contract changes are
individually compatible and so produce no failure at all: relaxing required or
cli_required, clearing required_when, widening enum, clearing interface_type,
and clearing property through a reviewed mapping exclusion. Any of those could
have ridden along with a reviewed type migration, leaving the exemption wider
than both its documentation and what the entry actually reviewed.
Replace the failure-list heuristic with a real equality check over every
published field except Type. Comparing the struct also means a field added to
parameterSchema later is covered automatically, instead of silently widening
every existing entry. The type check moves back ahead of the field loop because
it no longer needs to observe the other findings.
Add a rejection case for each individually-compatible direction. Each case first
asserts that the drift alone really is compatible, so it keeps exercising the
equality guard instead of quietly duplicating one of the incompatible-bundle
cases.
Verified against the previous implementation: with the old guard all six new
cases fail while the nine incompatible-bundle cases still pass, which is exactly
the gap that was reported.
Address the two P1 review findings and the coverage-gate CI failures:
- Every install/upgrade path that removes a skill dir (opposite-mode
leftovers, stale dingtalk-* / dws-shared, and same-name refreshes) now
moves the directory to ~/.dws/skill-backups/<stamp>/ first across
install.sh, install-skills.sh, install.ps1, install.js, `dws skill
setup`, and `dws upgrade`. A backup failure preserves the original
directory and never removes it.
- Remove --yes from every copyable `dws skill setup` example and document
what the command may remove; add regression tests that declining the
confirmation performs no removal and that the confirmation previews
every directory slated for backup+removal.
- Rename the skill-mode tests to the TestCrossPlatformCoverage* prefix so
the platform coverage gate selects them, and add edge tests for the
backup/prune/cleanup fallback branches, restoring changed-code coverage
to 100%.
Co-authored-by: Cursor <cursoragent@cursor.com>
schema-compatibility is the third check in the same Interface Integrity job,
after the two CLI interface gates. It also rejected every published parameter
type change outright. Because the earlier gates failed first and `set -e`
stopped the step from ever running, this one never surfaced in CI, so the
previous exemption only covered two thirds of the problem.
checkParameterCompatibility now consults a precise allowlist: the tool path,
parameter name and both type values must match exactly, making it
direction-sensitive by construction, and it applies only when nothing else the
gate checks about the parameter moved (default, interface_default, format,
property, interface_type, required, cli_required, required_when, enum). The type
check moved to the end of the function so the carve-out can see those findings;
ordering is unobservable because the result is sorted.
The only entry is "minutes/minutes.apply_minutes_permission" parameter "policy"
migrating from "string" to "integer" (for #912). That type is projected from the
Cobra flag type (provenance cobra_flag_type), so it describes how the CLI accepts
a value. Consumers build a command line from it, and "--policy 4" is the same
argv under either declaration — a quoted "--policy \"4\"" still reaches pflag as
4 — while RunE keeps enforcing the same [2,4] domain. The parameter maps to
property "policyId", which the command has always sent as a number, so "integer"
is closer to the actual request than "string" was.
Table values must be the canonical form schemaType emits: the JSON encoding of
the type keyword, so `"string"` with its quotes rather than a bare string. The
guard test recomputes both through schemaType and checks the decoded name
against the closed JSON Schema type set — reviewedInterfaceRefRedirect was
silently disabled twice by exactly this class of spelling mistake.
mergedFlagContractOtherwiseChanged was only ever exercised on the path where
every condition holds still, because `||` short-circuits: with no reviewed
entry the first operand already decides the outcome and the function is never
called at all. That left its five regression branches uncovered and put
changed-code coverage at 88.0952% against a 100% target.
Add the merge-path counterpart of the checkCompatibility bundled-regression
table, pairing each of shorthand / required / hidden / no-opt / scope with the
reviewed type change and requiring the type failure to reappear. Changed-code
coverage is now 100%.
The authoritative interface baseline and command-compatibility gates
rejected every flag type change on a historical command, with no review
channel — even when the new type only moves the same validation from RunE
to flag parsing. Both now consult a precise allowlist.
An entry must match command path, flag name and both type names exactly,
so it is direction-sensitive by construction, and it applies only when
nothing else about the flag moved (shorthand, required, hidden, no-opt,
scope). A bundled regression re-reports the type change.
The first and only entry is "dws minutes permission apply --policy" moving
from string to int (for #912): the old RunE parsed with
strconv.ParseInt(v, 10, 64) and enforced [2,4], the new one lets pflag
parse with base 0 and still enforces [2,4], so the historical set of
successful invocations is a subset of the new one. Base 0 additionally
accepts spellings like "0x3", which widens rather than narrows. Defaults
are excluded from the guard because the migration necessarily changes one.
In the snapshot gate the exemption resolves against the canonical
Command.Path, never the alias-expanded accepted path: an aliased command is
compared once per accepted spelling, so keying on that would let every
alias bypass the table.
The table is duplicated because check-authoritative-interface-baselines.sh
copies the whole scripts/policy/interface-baseline directory into a
worktree checked out at a historical revision and builds it there, so that
copy cannot import a package this branch adds. A guard test fails if the
two copies drift.
Upstream reorganized the multi-skill layout (#887: long-tail skills folded
into dingtalk-misc, dws-shared renamed to dingtalk-shared). Conflict
resolution keeps this branch's multi-by-default semantics (install.sh /
install.ps1 / skill setup default to multi; interactive prompts list multi
first) and adapts the cleanup paths to the rename: cleanup predicates now
recognize both dingtalk-shared (new bundle name, covered by the dingtalk-
prefix) and the legacy dws-shared so full installs and mode switches remove
pre-rename leftovers.
Co-authored-by: Cursor <cursoragent@cursor.com>
--ranges validated the position of "!" in the raw string and then returned the
trimmed halves, so " !A1:B2" was accepted and produced a set_cell_range /
clear_range operation carrying sheetId: "". Depending on how the server treats
an empty sheetId, the whole batch_update fails, or — worse — the operation lands
on the default worksheet instead of the one the user named, while the command
reports success.
Both halves must now be non-empty *after* trimming. batch-clear grew the same
hole independently (it duplicated the split inline); it now shares
splitSheetPrefixedRange, so the invariant holds by construction rather than by
being repeated correctly in two places.
batch-set-style --batch had the same gap at the JSON level: it only rejected
sheetId == "", so " " passed. It now judges the trimmed value but still sends
the raw one — sheetId may be a worksheet *name*, and names may legitimately
carry leading or trailing spaces, so trimming on the user's behalf would target
a different sheet. The --ranges form cannot express such a name anyway, which is
what --batch is for.
TestBlankSheetIdentifierIsRejectedBeforeAnyRemoteCall covers all three entry
points with calls == 0; TestBatchStyleSheetIDIsSentVerbatimNotTrimmed pins the
no-normalisation half.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Both capabilities were added as flags on an existing leaf, and in both cases
the leaf's published interface stopped describing what the command did:
- `sheet create --values/--sheets/--styles` orchestrates create → probe →
resolve default worksheet → write → read back → optional styles, yet the
leaf still published `interface_mode: mcp` + `create_workspace_sheet`.
- `sheet export --export-format csv` reads `get_range_as_csv` and never
invokes `submit_export_job`, yet the leaf published `submit_export_job`.
Each moves to its own command, declaring the interface it actually uses:
`sheet create-with-data` is `composite` with a reviewed reason and no
`interface_ref`; `sheet export-csv` is `mcp` + `get_range_as_csv`. Both are
pinned in the interface-disposition contract test.
`sheet create` and `sheet export` are restored byte-for-byte to main, so the
compatibility gates see two `command_added` additions instead of four
locked-field changes. The split also removes a user-visible trap: `--range`
without `--export-format csv` used to be silently discarded and the whole
workbook exported; the cross-format flags no longer exist, pinned by
TestSheetExportAndExportCsvFlagsDoNotLeak.
Drops the 8 now-stale mapping-ledger exclusions that covered the flags on
`sheet.create_workspace_sheet` / `sheet.submit_export_job`, shrinking this
branch's exclusion surface. Skill references and CHANGELOG follow the split.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
parseBorderStyles read only style and a string color, so every other key and
any non-string color was silently dropped:
{"top":{"style":"solid","colour":"#f00"}} succeeded and drew a border with no
colour, and color: 123 did the same. That contradicts the unknown-key
rejection this PR applies to --sheets and --styles — a partially applied
style reported as success is harder to notice than an error.
Each edge now accepts only style/color, rejects near-miss spellings with the
canonical key, and fails when style or color is present with the wrong type
or empty. All three entry points (set-style --border-styles-json,
batch-set-style --ranges/--batch, and create --styles border_styles) share
parseBorderStyles, so one fix covers them; tests assert the rejection happens
before any MCP call on every path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sweeping the same class the last review round hit: limits and behaviour this
PR added that only reached the Go long help, not the skill references an
agent actually reads.
- batch-set-style: the 200000-cell cumulative cap across all ranges was
missing (the 100-range cap and the atomic rollback were already there).
- sheet create --styles: size must be a positive integer (a fraction is
rejected rather than silently truncated) and the row/column range forms
reject trailing characters.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
json.Decoder.Decode returns after one value, so `--values '[[1]] trailing'`
was accepted as a valid matrix and the document got created anyway. A paste
that ran long, leftover shell concatenation, or two JSON values glued
together would silently drop the tail and still create a document the user
never asked for — and creation cannot be rolled back atomically. Require
EOF after the first value on both flags, following decodeOARequest.
docs(sheet): document the fail-closed CSV export and --allow-truncated
The skill references still claimed an oversized table is truncated with a
warning on stderr, and omitted the flag. The command now aborts before
writing anything when the server reports hasMore, so an agent relying on the
skill would misread the result and had no way to learn how to opt in.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
sheet create promises that every structural check happens before the first
MCP request, but each sheet spec was only checked for object type and name:
columns/data/dtypes/formats/startCell were left to table_put, so a bad type
created and renamed the remote document first and failed at write time,
leaving behind a document the user never successfully asked for. Validate
every provided field against table_put's input contract up front, and reject
the malformed {"sheets":"bad"} wrapper instead of treating it as one spec.
Also fixed while auditing the same flow:
- unknown/misspelled keys are now rejected in both --sheets and --styles.
The server DTOs are fixed beans, so a stray "datas" was silently dropped
and the read-back probe landed on the header row: full data loss reported
as success. Near-miss spellings get the canonical key in the message.
- columns is required (the server requires it), non-blank and trim-unique;
dtypes/formats keys must resolve to a column, since the server looks them
up by trimmed name and silently ignores the rest.
- sheetId inside a spec is rejected: the document does not exist yet.
- the read-back probe now honours header:false, mode:append (a fresh sheet
appends at row 1, the startCell row is ignored) and $-absolute/lowercase
startCell refs, which the server accepts after uppercasing.
- --values cells must be scalars; a map used to be written as "map[a:1]".
- the 30000-cell and 2000000-char write limits are enforced locally.
Docs: the --styles top level only accepts snake_case (camelCase aliases are
inner-field only), and the read-back probe is not pinned to A1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The --export-format csv branch reads get_range_as_csv, but the sheet export leaf
still declares interface_ref: submit_export_job, so discovering the csv
capability through Schema yields the wrong backing interface. Audit metadata
only — interface_ref is not read at runtime and routing is unaffected. Recorded
here so the limitation reaches release notes rather than living only in a code
comment; accurate attribution is tracked as follow-up.
The allowlist added in the previous commit keyed the reviewed
sheet.range_set_style migration by bare RPC name ("update_range"), but
interface_ref holds the canonicalized JSON that parseTool produces via
canonicalRawJSON. The lookup therefore never matched, the carve-out was
effectively disabled, and the real gate failed with
`schema tool "sheet/sheet.range_set_style" changed interface_ref`.
The existing redirect test did not catch this: it registered the fixture entry
using its own value and then asserted with the same value, so any format would
have passed. That is the same mistake as keying a probe on the author's field
spelling instead of the wire contract.
- The allowlist entry now uses the compact canonical JSON, taken from the gate's
own output rather than from pretty-printed `dws schema`.
- TestCrossPlatformCoverageReviewedRedirectKeysAreCanonicalJSON recomputes every
registered key through canonicalRawJSON, so a bare name or a pretty-printed
variant fails locally instead of only in CI.
sheet export keeps its reviewed mcp + submit_export_job declaration. The comment
now records the known trade-off explicitly: --export-format csv is a mutually
exclusive branch that reads get_range_as_csv and never invokes the declared
submit_export_job, so this declaration does not cover the csv branch's backing
interface. Attributing that branch is left out of scope for this change.
Verified against the real gate, not just unit tests: all four Interface
Integrity checks pass (authoritative-interface-integrity,
check-command-compatibility, schema-compatibility, skill-command-integrity).
Two review findings, plus a same-class defect found by sweeping for it.
1. compatibleInterfaceRefRedirect accepted any mcp tool repointing from any
non-empty interface_ref to any other, as long as no other check for that tool
failed. Schema shape cannot prove two RPCs share business semantics,
permissions, error behaviour, or side effects, so that would have let every
future backend swap bypass the gate it exists to enforce. It is now keyed on
an explicit reviewedInterfaceRefRedirect allowlist of exact tool + old→new
pairs, currently holding only the reviewed
sheet.range_set_style: update_range → set_cell_range migration. Every other
ref change is reported again.
2. sheet export --export-format csv only printed a stderr warning when
get_range_as_csv returned hasMore=true, then wrote --output and reported
success with exit code 0. Automated callers, and anyone not watching stderr,
would treat an incomplete file as a complete export, and an existing target
file was overwritten with truncated data. Truncation now fails before the
write (leaving any existing file untouched) unless --allow-truncated is
passed; with the opt-in the success line states the data is incomplete.
--allow-truncated is registered in the reviewed mapping ledger as a local
policy input.
3. Swept for the same classes and found sheet_dimension.go repeating the
fmt.Sscanf("%d") prefix-parse hole in three places: insert_dimension,
delete_dimension, and update_dimension all accepted --length "3x" as 3, so a
malformed value silently operated on the wrong row/column count — the delete
direction is not rollbackable. All three now use strconv.Atoi. (Checked and
cleared: sheet csv-get also surfaces hasMore, but it has no --output and only
returns the flag in its JSON payload, so it is not the same fail-open shape.)
Tests: allowlist rejection cases (unreviewed target ref, and the same pair on a
tool absent from the allowlist, asserted outside the table so the registration
survives until checkCompatibility runs); truncation fail-closed with a
pre-existing output file asserted byte-for-byte unchanged; --allow-truncated
write-through; and an untruncated read needing no opt-in. Changed-code coverage
stays at 100% (939 statements).
parseRowColRange gates --styles row_sizes/col_sizes before the document is
created. The row branch parsed with fmt.Sscanf(a, "%d", &r1), which consumes
only the leading digits and does not require the whole token, so "1x:3" was
silently accepted as row 1 and "2foo" as row 2. Such input passed pre-flight,
then update_dimension was sent to the wrong row after the document and data had
already been created — an unrollbackable wrong edit, the opposite of the gate's
purpose. The row branch now parses with strconv.Atoi, which requires the entire
token to be a valid integer.
The column branch had the same class of hole via a different path: parseA1Cell
appends "1" to the column token, so "A5" became "A51" and was accepted as
column A. A new isAllLetters pre-check requires the column token to be non-empty
and letters-only before parsing; genuine multi-letter columns like "AX" still
pass. With that guarantee the subsequent parseA1Cell can no longer fail, so its
now-dead error branch is removed.
Adds trailing-character rejection cases to TestParseRowColRange: "1x:3",
"2foo", "1 2:3" (rows), "A5:C", "A1", ":C" (columns), plus "AX:C" to confirm
multi-letter columns remain valid. Changed-code coverage stays at 100%.
Three review P1s.
1. CHANGELOG no longer asserts a breaking Schema change this PR does not ship.
sheet export / sheet create deliver main's mcp + interface_ref and
schema-compatibility reports ok (0 changed fields), so the "declared as
composite (breaking)" entry was false and is removed; the set-style entry
drops the "breaking" framing (accepted as compatible by the reviewed
mapping-exclusion carve-out); the duplicate ### Changed heading is merged.
2. firstNonEmptySheetSpecCell reads the start cell via
pickStr(spec, "startCell", "start_cell"). Sheet specs are forwarded verbatim
to table_put, whose wire fields are camelCase in this repo. The prior
snake_case-only read meant a user passing the real startCell would have data
written at the offset while the probe read A1 — a false "写入未生效" on a
successful write. camelCase preferred, snake_case kept for tolerance.
3. planStyleOps now parses cell_merges range with parseA1Range, matching the
cell_styles branch. planStyleOps is dry-run once before create_workspace_sheet
as the up-front structural gate, so an invalid range like "not-a-range" is now
rejected before any RPC instead of failing only at the final merge_cells call
and leaving an unrollbackable partially-completed document. merge_cells' range
contract is A1:B3-style, which parseA1Range covers (and it strips a Sheet1!
prefix).
Tests: cell-merges-invalid-range added to TestSheetCreateValidatesBeforeCreating
Document (asserts calls == 0); TestFirstNonEmptySheetSpecCell covers both
startCell and start_cell. Changed-code coverage stays at 100%; targeted sheet
suites pass; CHANGELOG has a single Changed section with no false breaking claim.
Adds coverage for the branches introduced by the per-sheet read-back:
resolveSheetIDsByName's RPC-error and unparseable-response paths, the create
--sheets path surfacing a list-fetch failure with the nodeId, and the
single-value data row in sheetSpecGrid. Changed-code coverage back to 100%.
The --values branch already reads back its first non-empty cell after writing,
to defend against the new-document initialization race where a write returns
success but the data does not land. The --sheets branch called table_put and
reported success with no read-back, so the same race would let the command exit
successfully while one or more sheets silently lost their initial data.
After table_put, the --sheets branch now:
- re-fetches get_all_sheets to build a name -> sheetId map (table_put reuses the
renamed default sheet and auto-creates the rest by name), and
- for every spec that actually has content, reads back its first expected
non-empty cell and fails if the read-back is empty or the sheet is missing.
firstNonEmptySheetSpecCell mirrors firstNonEmptyValuesCell: it treats columns as
the header row followed by data rows, honours start_cell, and returns
hasContent=false for a name-only spec so a legitimately empty sheet is not
misreported as data loss. Failures carry the nodeId and point at
sheet table-put for recovery, matching the --values branch's error shape.
Tests:
- TestSheetCreateWithSheetsVerifiesEachSheetLanded covers an empty read-back
(errors, naming the sheet + nodeId + table-put), a sheet missing from the
post-write listing, and a name-only sheet that must not trigger a read-back.
- TestFirstNonEmptySheetSpecCell covers header/data origins, an empty first
header cell, a start_cell offset, and content-less specs.
- Existing --sheets tests updated for the added get_all_sheets + per-sheet
read-back calls.
Reverts the interface_mode of sheet create and sheet export from composite back
to mcp with a single interface_ref, matching upstream/main and the existing
convention for multi-tool leaves (doc.create_document declares mcp +
create_document even though it also calls update_document).
Rationale:
- interface_ref is audit / traceability metadata; nothing reads it at runtime
(verified: rebuilding with a bogus interface_ref still routes to the correct
tool). Declaring the primary tool and treating the orchestration as an
implementation detail is the pattern main already uses.
- sheet export was mcp + submit_export_job on main; it already orchestrated
submit_export_job + query_export_job without declaring the poll. Adding an
--export-format csv branch does not change that shape, so it does not warrant
flipping to composite.
- sheet create was a genuine single-RPC command on main (create_workspace_sheet
only). The --values / --sheets / --styles orchestration I added runs after the
document exists; per the doc.create_document precedent it stays mcp.
This takes the sheet schema-compatibility failures from 4 to 0 without a waiver
or an admin override: the declarations now equal main's.
Also updates the wording of the seven new mapping-exclusion reasons for these
two commands (submit_export_job CSV params, create_workspace_sheet
values/sheets/styles) from "Composite ... input" to "Wrapper ... input", so the
reason text no longer collides with the interface_mode value now that both
leaves are mcp. The reasons are otherwise unchanged and still describe where
each value actually goes. range_batch_set_style keeps its "Composite" wording
because it genuinely stays interface_mode=composite.
Removes the two composite entries for these leaves from the interface
disposition contract test.
No execution path changes; targeted sheet suites, the disposition contract test,
and the schema-compat policy tests all pass; check-schema-catalog is green;
sheet-scoped schema-compatibility reports 0 failures.
Two compatibility carve-outs, written alongside the existing interface_type
retirement allowance. Both cover declarative provenance metadata that nothing
reads at runtime: the tool a leaf invokes is decided in the CLI source, so a
stale interface_ref or property misinforms a reader rather than misrouting a
call. Neither carve-out can mask a change to the surface callers depend on.
Cleared property through a reviewed mapping exclusion. A leaf whose backing RPC
moves to a nested payload has no honest flat property to publish. The two
alternatives are worse: keep naming a field the request no longer contains, or
let assembly fall back to flag_name_inference and publish a name that appears in
no request at all. Accepted only when the old value was non-empty, the new value
is empty, and the new value resolved through reviewed_mapping_exclusion. A
redirect to a different non-empty value, a clearing by inference or native
annotation, a clearing with no recorded source, and populating a previously
empty property all stay incompatible. The exclusion table cannot be abused to
wave arbitrary clearing through: internal/cli/schema_parameter_bindings.go
verifies every parameter claiming an exclusion really does deliver an empty
property, and every entry carries a non-empty reviewed reason.
Redirected interface_ref with an unchanged CLI contract. Accepted only when
interface_mode is unchanged and stays mcp, both refs are non-empty, and no other
compatibility failure was recorded for that tool. That last condition is the
operative definition of "the contract is unchanged" — it is measured, not
asserted, so it automatically covers a lost parameter, a newly required one, a
moved type / default / format / enum, a tightened constraint, a positional or
dry_run change, and any effect / risk / confirmation / idempotency move. Any one
of them re-reports the redirect, so a surface change cannot ride along behind a
backend move. Moving to or from composite is a change in kind rather than a
redirect and stays reported; so does removing a ref outright.
Deliberately still incompatible: mcp -> composite with the ref dropped. That is
a leaf declaring it now orchestrates several RPCs, which is a semantic upgrade
rather than a like-for-like substitution, and it belongs in review.
For this branch the two carve-outs take schema-compatibility from 17 changed
fields to 4 — the twelve sheet.range_set_style property clearings and its
update_range -> set_cell_range redirect are now accepted. The remaining four are
the interface_mode plus interface_ref pairs on sheet.create_workspace_sheet and
sheet.submit_export_job.
Tests: TestCrossPlatformCoverageSchemaCompatPropertyClearingExclusion and
TestCrossPlatformCoverageSchemaCompatInterfaceRefRedirect assert the accepted
shapes plus twelve neighbouring shapes that must stay incompatible; the drift
table gains cases for clearing without an exclusion and redirecting despite one.
P1 from automated review. --values decoded with plain json.Unmarshal, so every
number became a float64 and integers beyond 2^53 were rounded before anything
was written. The read-back only checks that the probe cell is non-empty, so the
corruption was reported as a successful write. Order numbers and snowflake IDs
are ordinary spreadsheet data.
Measured before the fix:
1234567890123456789 -> 1234567890123456768 snowflake id, tail rewritten
12345678901234567890 -> 12345678901234567000 20-digit order number
9007199254740993 -> 9007199254740992 2^53+1
--sheets had the same defect, which the review did not mention: its records and
data are forwarded verbatim to table_put, and the float64 round trip rewrote
1234567890123456789 as 1234567890123456800 before the request left the CLI.
Both channels now decode with json.Decoder.UseNumber, and cellToString emits a
json.Number through its String method so no float conversion happens on the way
to CSV. --styles keeps plain Unmarshal on purpose: its numbers are font sizes and
pixel dimensions, already constrained to int32 by pickNum, with no large-integer
case.
Tests assert the payload the CLI actually sends, not a recomputed decode:
- TestSheetCreatePreservesLargeIntegerLiterals checks the csv argument of
set_range_from_csv and the marshalled table_put arguments. Both halves also
assert the rounded forms are absent, so removing UseNumber fails the test
instead of passing on a lucky substring match.
- TestCellToStringKeepsJSONNumberVerbatim covers large, negative, fractional and
exponent literals, and keeps the existing float64 behaviour for other callers.
The shared scriptedToolCaller keeps only the last call, and the write is the
fourth of five, so the assertion needs an intermediate call. Rather than extend
that shared helper, this adds a callRecorder local to this file: InitDeps takes
the edition.ToolCaller interface, so embedding *scriptedToolCaller and
overriding CallTool is enough.
Changed-code coverage stays at 100.0000% (850 statements) per
check-coverage-gate.sh --changed-only.
Two things, both in the create-with-data path.
Unique worksheet names. parseCreateSheetSpecs accepted a --sheets payload with
repeated names, so table_put created several worksheets sharing one name. The
style tools locate a worksheet by "id or name", so --styles would then land on
whichever duplicate the server picked, and --styles runs after the document
already exists and cannot be rolled back. Duplicates are now refused before
anything is created, naming the first occurrence:
--sheets[1].name="一月" 与 --sheets[0] 重复；工作表名必须唯一，...
Case-only differences are still accepted: the server distinguishes them and the
CLI should not tighten that. --styles needs no equivalent check because it
already requires name equality with the corresponding --sheets entry, and those
are now unique.
Coverage. The previous commit added a precise pickNum error path to the standard
and auto branches of planSizes, but no test reached it: the existing cases used
an integer size, which stops at "不能同时给 size" before the numeric check runs.
The CI coverage gate therefore reported 99.7619% on changed code
(sheet_create_with_data.go:512-514 and :521-523). Two cases now drive
type=standard and type=auto with size 28.5.
That pair is not only about the percentage. It pins the error precedence: a
fractional size must report "size=28.5 必须是整数", which points at the field
actually written wrong, rather than the generic "不能同时给 size". Swapping the
two checks would make the message misleading and now fails the tests.
Changed-code coverage measured locally at 100.0000% (845/845 statements), with
no uncovered blocks in the diff against upstream/main.
Tests:
- TestParseCreateSheetSpecsRejectsDuplicateNames covers adjacent, non-adjacent
and {"sheets":[...]}-wrapped duplicates, plus three payloads that must pass.
- Three new cases in TestSheetCreateValidatesBeforeCreatingDocument
(sheets-duplicate-name and the two fractional sizes), each asserting calls == 0.
P2 from automated review. pickNum ran int(n) straight on the float64 that JSON
decoding always produces, so font_size: 12.9 and row_sizes.size: 28.5 were
silently rewritten to 12 and 28 and then executed as a valid configuration. Both
the help text and the error messages state these fields must be positive
integers, and --styles is a non-atomic sequence that cannot be rolled back, so
truncation left a sheet that did not match what the caller asked for.
Measured before the fix:
font_size=12.9 -> emitted fontSize=12
size=28.5 -> emitted pixelSize=28
size=1e20 -> emitted pixelSize=9223372036854775807
The overflow case was worse than reported: int(1e20) saturates to MaxInt64 and
was still sent.
pickNum now returns an error and rejects a non-integral value, a NaN or Inf, a
magnitude outside int32, and a non-numeric type. A missing key and an explicit
null still report "not provided" without an error, so optional fields keep
working. Every call site propagates the error, which means the whole --styles
payload is refused before the document is created. Confirmed through the real
CLI: font_size=12.9, col_sizes size=120.5 and size=1e20 all fail with a specific
message while font_size=12 still passes.
Tests:
- TestPickNumRejectsNonIntegralAndOutOfRange covers eight rejected inputs plus
five accepted ones, the missing key, an explicit null, and alias-key lookup.
- Four new cases in TestSheetCreateValidatesBeforeCreatingDocument for
cell_styles font_size, row_sizes size, col_sizes size and the overflow, each
asserting calls == 0.
- TestPickStrAndPickNum updated: a bool value now surfaces a type error with
ok=true rather than being reported as absent.
Adds four capabilities and, for the style surface, moves to the interface that
can actually express them.
Added:
- sheet create --values / --sheets / --styles: create a workbook and populate it
in one command. --values takes a 2D array into the default sheet, --sheets
takes typed tables across several sheets, --styles carries cell_styles /
row_sizes / col_sizes / cell_merges. Every structure and enum is validated
before the document is created, so an invalid config never leaves an orphan
empty document behind.
- sheet export --export-format csv: synchronous single-sheet RFC4180 export with
--sheet-id, --range and --value-render-option. --output writes to a file (a
directory gets sheet-export.csv), otherwise the CSV goes to stdout while the
truncation warning goes to stderr, keeping stdout pipeable.
- sheet update-dimension --size-type: pixel / standard (restore the default row
height or column width) / auto (fit row height to content, ROWS only).
- sheet replace --match-formula: search and replace inside formula text.
- sheet range set-style --font-style / --font-line / --font-family /
--border-styles-json.
- sheet range batch-set-style --ranges: stamp one style across several
sheet-qualified ranges.
Changed (breaking Schema change, no CLI break):
- sheet range set-style moves from update_range to set_cell_range. The
update_range style channel exposes exactly eight properties
(backgroundColors, fontSizes, horizontalAlignments, verticalAlignments,
fontColors, fontWeights, wordWrap, numberFormat) and has no slot for italic,
underline/line-through, font family or borders, so the four new dimensions are
not expressible there. interface_ref becomes set_cell_range and the twelve
style flags stop publishing a flat property, because the value now lands in
cells[i][j].cellStyles.* with no single top-level field to name.
- sheet range batch-set-style submits one atomic batch_update instead of looping
update_range, so a partial failure no longer leaves half the ranges stamped.
--continue-on-error becomes a server passthrough. Caps the fan-out at 100
ranges and 200000 cells in aggregate.
- sheet export and sheet create declare interface_mode=composite: both route
across several tools depending on the flags, so a single mcp ref was wrong.
Schema hygiene:
- Nineteen parameters that previously resolved through flag_name_inference into
property names present in no request (bgColor, exportFormat, values, ...) are
now reviewed mapping exclusions with a stated reason, so Schema omits the
property instead of inventing one.
schema-compatibility reports 17 changed fields: 13 on sheet.range_set_style
(interface_ref plus twelve property mappings) and 2 each on
sheet.submit_export_job and sheet.create_workspace_sheet (interface_mode plus
interface_ref). No CLI flag is removed and no command path changes; the same
invocation runs on both the old and the new binary. Landing this needs a
decision on the Schema contract break.
Tests: targeted sheet suites in internal/helpers and the interface disposition
contract tests in internal/app pass; make build and gofmt clean;
check-schema-catalog, check-generated-drift, check-command-surface,
check-skill-commands and check-runtime-confirmation-truth all pass.
Replace the oversized mono Sheet reference with the progressive routing layout, mirror all Sheet topic references across both bundles, and add a paired-tree drift guard.
Update Long and --yes flag text to match the upfront confirmation gate
and automatic sign retry after --yes, consistent with thread trash.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add explicit confirmation_required gate before any share_message_to_chat
call so piped stdin or direct server success cannot bypass --yes. Keep
sign retry after confirmation and add regression tests for zero-call deny,
sign retry, and direct-success paths.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep schema-compat tests aligned with main after removing the ding
property-correction allowlist; the hard-fail case is already covered elsewhere.
Co-authored-by: Cursor <cursoragent@cursor.com>
Hit the nil-allowlist early return so the platform changed-statement gate
reaches 100% after shared-file overall baseline filtering.
Co-authored-by: Cursor <cursoragent@cursor.com>
Merge of the Deprecated cache refresh surface reintroduced a public root
command; refresh the CLI interface baseline so cli-smoke stays green.
Co-authored-by: Cursor <cursoragent@cursor.com>
Deleting 100%-covered packages such as recovery falsely regressed overall
percent against merge-base. Baseline overall now uses only files still
present in the candidate profile, with a 0.1pp CI tolerance. Also exercise
stripSchemaValueCompact nested map/slice branches.
Co-authored-by: Cursor <cursoragent@cursor.com>
Restore dws cache {refresh,status,clean} as a successful no-op
compat surface so historical scripts/agents keep working after
static-endpoint delivery. Deprecated leaves stay out of Schema via
IsAvailableCommand without schema exclusions.
Co-authored-by: Cursor <cursoragent@cursor.com>
Rewrite incomplete `dws devapp +` and non-product `dws finance` mentions
so skill-command-integrity no longer treats them as executable paths.
Co-authored-by: Cursor <cursoragent@cursor.com>
Visible Deprecated recovery stubs are part of the public root command
tree; refresh the CLI interface baseline so cli-smoke stays green.
Co-authored-by: Cursor <cursoragent@cursor.com>
Cover each equals-form flag in its own switch case so the platform
changed-statement coverage gate reliably hits both branches.
Co-authored-by: Cursor <cursoragent@cursor.com>
Deprecated recovery leaves are not public schema leaves
(IsAvailableCommand=false), so listing them as exclusions fails
completeness with stale exclusions.
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop Hidden so authoritative interface integrity still passes, restore
the CI historical command gate, and keep the unsupported notice without
Skill guidance.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep a Hidden compatibility shim that returns 不再支持 for plan/
execute/finalize, so this release stops supporting the surface while a
later release can delete the shim entirely.
Co-authored-by: Cursor <cursoragent@cursor.com>
Allow intentional removal of public commands (e.g. dws recovery)
without compatibility stubs. Keep Schema and skill-command checks
in the Interface Integrity job.
Co-authored-by: Cursor <cursoragent@cursor.com>
- introduce removeChatToolbarCustomShortcutFn in toolbar_remove_custom.go
as a package-level injection seam; default impl routes through
callMCPToolOnServer against the im server so production behavior is
unchanged
- update RunE to dispatch via the seam instead of calling
callMCPToolOnServer inline
- add two TestCrossPlatformCoverage* tests that swap the seam via
testseam.Swap and verify: (1) without --yes the seam is never called
and a typed confirmation_required error is returned, (2) with --yes
the seam is called exactly once with openCid and shortcutId. The
stub forwards to deps.Caller.CallTool so the user_required contract
gate (leaf.go) still sees the CallTool channel.
Remove EXPERIMENTAL/Preview banners from skill setup, install scripts, README, and misc references so multi mode is no longer framed as unstable preview.
Co-authored-by: Cursor <cursoragent@cursor.com>
Do not expand the default policy gate with mono-multi or skill-commands;
leave them as optional make targets only.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Drop --yes and shell-comment example lines from the Cobra Example
field in chat toolbar remove-custom; keep only the single
non-bypassing command line. Aligns with AGENTS.md "no --yes in
stored examples" and "No shell comments in examples" rules.
- Mirror the change in skills/mono/references/products/chat.md
toolbar remove-custom block: remove the duplicated --yes and
shell-comment lines; keep Flags block and prose note untouched.
- Add two end-to-end confirmation gate tests under
internal/helpers/toolbar_helpers_test.go using the existing
toolbarTestCaller seam (extended with a calls []toolbarCall
slice so the new tests can assert call counts as well as the
most recent call):
* TestCrossPlatformCoverageToolbarRemoveCustomRejectsWithoutYes
asserts confirmation_required and zero MCP calls when --yes
is omitted.
* TestCrossPlatformCoverageToolbarRemoveCustomCallsMCPWithExactArgsWhenYes
asserts exactly one im/remove_chat_toolbar_custom_shortcut
call with openCid=<cid> and shortcutId=<id> when --yes is
set.
- No changes to Contract.Selection.Examples (already compliant),
Long prose, or any other toolbar file. Helper field addition is
additive: legacy single-call fields stay so all prior tests
remain green.
Fixes: PR #877 CR P1 (remove-custom confirmation gate).
Risk tier: Standard.
Verification: see PR description.
Prefer schema --compact in agent docs, remove服务发现/cache teaching,
and stop doctor from reporting a no-op cache health item.
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop the no-op dws cache stubs and the doctor cache health item, and
scrub skill/AGENTS guidance that still pointed agents at them.
Co-authored-by: Cursor <cursoragent@cursor.com>
- add 45 public document shortcuts and 2 reviewed expert-only paths
- preserve six historical command and Schema identities alongside canonical leaves
- add safe local download primitives and document access/share orchestration
- keep comment create/reply confirmation backward-compatible
- ensure grant-and-share upgrades insufficient roles before messaging
- return non-zero partial/failure message ledgers and structured partial-write recovery metadata
- enumerate every selection candidate and use rune-safe Unicode keyword contexts
- assert zero-call confirmation boundaries for destructive shortcuts
Validation:
- full Go test suite and repository policy
- real DingTalk E2E for 34 canonical shortcuts, all 8 compatibility-affected entries, READER-to-EDITOR grant-and-share upgrade, and same-block selection ambiguity with zero comment writes
- command compatibility across 1,221 historical nodes and complete Schema compatibility
- 1,152 Agent examples including 62 real Cobra dry-runs
- 100% changed-code coverage across 1,260 executable statements
Complete ding leaf Property/Required from Execute CallMCP keys and live
help semantics, and allowlist the four inference→declare property remaps
so schema-compat does not freeze wrong camelCase flag names.
Co-authored-by: Cursor <cursoragent@cursor.com>
Production already ships an empty pin; remove the leftover mcp_metadata
candidate path so parameter resolution only uses declare/Cobra sources.
Co-authored-by: Cursor <cursoragent@cursor.com>
Host hrbrain, markdown, pat, and profile under dingtalk-misc, retire
their standalone packages, and rename dws-shared to dingtalk-shared
across live paths, coverage, installers, and policy.
Co-authored-by: Cursor <cursoragent@cursor.com>
Point coverage, installers, IM skill-chain, and shared routing at
dingtalk-misc references after retiring the standalone event package.
Co-authored-by: Cursor <cursoragent@cursor.com>
Host personal IM event docs under misc like other long-tail products, and
retire the standalone multi skill package.
Co-authored-by: Cursor <cursoragent@cursor.com>
Point coverage, shortcut generation, installers, and shared routing at
dingtalk-misc references after retiring the standalone packages.
Co-authored-by: Cursor <cursoragent@cursor.com>
Host open-platform app docs and skill-market commands under misc like
other long-tail products, and retire the standalone multi skill packages.
Co-authored-by: Cursor <cursoragent@cursor.com>
Align mono/multi minutes/doc/conference skill facts with live CLI: prefer
--limit/--cursor, drop false participants claims, replace deprecated doc
search, and mark conference as unsupported without dead conference.md links.
Co-authored-by: Cursor <cursoragent@cursor.com>
Sync transport post-recovery hints into en/zh locales, refresh skill QA docs for Phase 1–3/4B, and remove the dead internal/recovery CI high-risk path.
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop the recovery package/commands and related Schema/skill teaching so agents
stop being steered at a dead surface, while keeping mono↔multi content QA work.
Co-authored-by: Cursor <cursoragent@cursor.com>
Specify coverage/structure/drift gates against mono, inventory existing
skill policy tests, and extend the §7 checklist for the QA track.
Co-authored-by: Cursor <cursoragent@cursor.com>
Defer install/upgrade behavior and cherry-picks to a follow-up branch;
keep this branch on multi/mono content layout and content contracts.
Co-authored-by: Cursor <cursoragent@cursor.com>
Limit scope to skill trees and skill install/setup/upgrade framework;
defer non-skill CLI, client pipelines, and full installer rewrites.
Co-authored-by: Cursor <cursoragent@cursor.com>
Capture inventory, port/adapt/reject decisions, and phased work before any
framework implementation on a main-based branch.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Remove --yes from Selection.Examples in toolbar_remove_custom.go
(schema_agent_examples.go forbids --yes in stored examples)
- Fix Confirmation "required" -> "user_required" in toolbar_remove_custom.go
(schema catalog requires enum value from {not_required, user_required})
- Fix Idempotency "not_idempotent" -> "non_idempotent" in toolbar_create_custom.go
(schema catalog requires enum value from {idempotent, non_idempotent, unknown})
Fixes: F1 BLOCK from stability-release-engineer round 5 review
- B1: Fix --sort-index 0 silent drop by using cmd.Flags().Changed()
instead of value comparison in create-custom and update-custom
- W1: Add MarkFlagRequired("shortcut-id") in remove-custom and
update-custom for consistent error messages
- W2: Extend SYSTEM_BUSY error handling to all write commands
(add/hide/create-custom/remove-custom/update-custom)
- W3: Add duplicate key detection in parseExtension to prevent
silent data loss on repeated --extension keys
Add 3 custom shortcut bar CRUD subcommands and update skill docs.
- toolbar_create_custom.go: create custom entry with extension parsing
and org-id-list support (write/medium, not_idempotent)
- toolbar_remove_custom.go: delete custom entry with --yes confirmation
gate (write/medium, confirmation required)
- toolbar_update_custom.go: update custom entry with same parameter set
as create-custom plus shortcut-id (write/medium)
- chat.md: add toolbar command group documentation with all 7 subcommands
Add `dws chat toolbar` command group with shared helpers and 4 basic
subcommands for managing conversation shortcut bar visibility and order.
- toolbar_helpers.go: shared utilities (hasIntersection, isSystemBusy,
parseExtension, toolbarConversationID, toolbarNewSystemBusyError)
- toolbar_helpers_test.go: unit tests for shared helpers
- toolbar.go: command group entry assembling 7 subcommands
- toolbar_list.go: list shortcut entries (read/low)
- toolbar_add.go: add entries to visible area (write/low)
- toolbar_hide.go: hide entries from visible area (write/low)
- toolbar_sort.go: sort entries with intersection validation and
SYSTEM_BUSY error handling (write/low)
- chat.go: mount newChatToolbarCommand() to chat root
After merging #861, use prepareWhiteboardCard and --yes so coverage
tests compile and match fail-closed confirmation + soft pending verify.
Co-authored-by: Cursor <cursoragent@cursor.com>
Capture runMarkdownUnifiedDiff/diffJSONMarshalIndent before spawning the
compute goroutine so testseam restores cannot race a late timeout path.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep the add-only Wukong compat flag, but publish it as a CLI-local
polymorphic dispatch without a download_file property binding.
Co-authored-by: Cursor <cursoragent@cursor.com>
Exercise markdown diff, mail export/share, drive latest/depth, whiteboard,
and diff-engine edges via TestCrossPlatformCoverage*; add small injectable
seams only where defensive branches are otherwise unreachable.
Co-authored-by: Cursor <cursoragent@cursor.com>
Expose minutes hot-word delete, permission apply, and audio-memo list from
live MCP gaps; add hidden/cross-product flag aliases only (never remove),
with TestCrossPlatformCoverage coverage for the new surfaces.
Co-authored-by: Cursor <cursoragent@cursor.com>
Whiteboard dry-run now stamps preview_kind=plan; mail export/share
drop [DRY-RUN] tags so plan evidence matches declared DryRunSpec.
Co-authored-by: Cursor <cursoragent@cursor.com>
Schema Manual examples forbid confirmation bypass via --yes; also stub
whiteboardSleep in product example coverage so race CI does not hang.
Co-authored-by: Cursor <cursoragent@cursor.com>
Align leftover multi skill banners with the non-experimental wording,
retarget markdown routing off dingtalk-misc, and remove the retired
SAFETY_PREAMBLE_INJECT marker plus the missing extract_media_id.py refs.
Co-authored-by: Cursor <cursoragent@cursor.com>
Port calendar event instances, markdown diff, drive list --latest,
mail calendar/calendar-event/shared-with-me/export/share-to-chat, and
doc whiteboard insert so open edition matches documented Wukong test
command surfaces.
Co-authored-by: Cursor <cursoragent@cursor.com>
Platform coverage only selects TestCrossPlatformCoverage*; name the new
declaration-only visibility regression accordingly and exercise the ForTest
deps restore helper.
Co-authored-by: Cursor <cursoragent@cursor.com>
Outside --dry-run, list_doc_versions uses the normal CallTool channel;
CallReadTool is reserved for the dry-run read path.
Co-authored-by: Cursor <cursoragent@cursor.com>
Declaration-only roots skip injectStaticServers; derive product visibility
from StaticServers directly so reverse completeness is not weakened, and
restore helpers deps via pointer snapshot instead of InitDeps(nil).
Co-authored-by: Cursor <cursoragent@cursor.com>
Homology Execute probes use NewSchemaSourceRootCommand (no InitDeps); skip
deps.Out when unset and InitDeps a throwaway caller on the probe path.
Co-authored-by: Cursor <cursoragent@cursor.com>
Schema assembly must mount the reviewed command tree without InitDeps or
SetDynamicServers, so a live process keeps its ToolCaller and plugin
endpoints. Also restore doc version revert dry-run short-circuit so
--dry-run skips remote version preflight.
Co-authored-by: Cursor <cursoragent@cursor.com>
Drop Windows-only DPAPI export/import names from the darwin -run filter;
those tests skip on macOS and already run under the Windows job.
Co-authored-by: Cursor <cursoragent@cursor.com>
TestLeafArgsOmitsEmptyAndNonPositive called BuildArgs with unset required
CSV flags; after rejecting empty required transforms that path correctly
errors. Satisfy required flags in the omit-empty case and cover the new
RequiredError / scalar-transform branches for the macOS coverage gate.
Co-authored-by: Cursor <cursoragent@cursor.com>
Separator-only inputs like --event-codes ',' passed pre-transform Required
checks, then BuildArgs omitted the key and ConfirmFirst write paths could
still call MCP. Enforce non-empty transform results for Required flags and
add CrossPlatformCoverage regression coverage.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep the darwin workflow contract from re-accepting the merged
keychain+auth+app invocation that main briefly required.
Co-authored-by: Cursor <cursoragent@cursor.com>
Port OA approval form-schema / forecast-process / create-instance from
main via DeclareLeafMetadata and mapping-ledger exclusions; keep retired
schema pin paths deleted and preserve the CI auth/keychain split.
Co-authored-by: Cursor <cursoragent@cursor.com>
Resolve the macOS race budget conflict in favour of the focused scope.
c1f96241 on main extended the whole-package macOS race step from 10m to
12m and pinned that budget in the workflow contract. This branch removes
the whole-package run instead: ./internal/app is already covered by the
Ubuntu "race: app" shard, and macOS only needs the natively-gated tests.
With the focused scope the job drops from 10m47s to ~3m, so the 12m
budget is no longer needed and the two step timeouts (6m + 5m) fit inside
the 15m job budget with headroom.
The contract assertion c1f96241 added is superseded rather than dropped:
pinning both focused commands locks the per-step timeouts, and the
existing checks still block a whole-package regression and require
./internal/app to appear exactly once.
Narrowing the macOS internal/app -run pattern orphaned
TestValidateNewBinary_RecoversFromUnsignedDarwin: it is the only
runtime.GOOS != "darwin" gated test in the package, the Ubuntu race shard
skips it on Linux, and the platform coverage gate only runs
^(TestAllShortcuts|TestCrossPlatformCoverage). No CI job selected it any
more, so it could never run or fail again.
- Add the self-heal test back to the macOS -run pattern.
- Add the (CrossPlatformCoverage)? group the Windows pattern already has,
which also recovers TestCrossPlatformCoverageAuthMigrateKeychainRemainingBranches.
- Attribute vacuous runs: the two skip paths now name the branch that went
unverified, and DWS_REQUIRE_AMFI_SELF_HEAL=1 escalates such a run to a
hard failure on a host that does enforce amfid. GitHub's hosted macOS
runners do not reproduce the amfid kill, so this test has been silently
skipping there all along.
- Restore step-timeout headroom: 10m + 5m exactly equalled the 15m job
budget, leaving none for setup. The keychain/auth step drops to 6m
(measured 2m43s).
- Add a contract test that couples the macOS -run pattern to the set of
darwin-gated tests in internal/app, so the next narrowing fails loudly
instead of silently orphaning one.
Cover applyInterfaceMetadataFallback success/audit paths and summary edges
that were lost when interface_metadata_test.go was retired, closing the
aggregate overall non-regression gap (~91.22% → above merge-base).
Co-authored-by: Cursor <cursoragent@cursor.com>
Unify natural target resolution and message contracts, add deterministic IM event listening, streamline cold-start skills, and cover the flows with schema gates and end-to-end tests.
- **Command typo guidance** — returns a validation error with up to three nearest command suggestions and the parent `--help` entry instead of printing the full command list.
- **Fork pull-request admission** — keeps the read-only Reviewer Router identity check fail-closed while allowing external contributors' CI to use the reviewed public App slug when GitHub withholds repository variables.
- **Markdown append chunking rewritten around safe split positions** — long markdown is now split so that every chunk is a complete, self-contained top-level block sequence, which is what `update_document mode=append` requires: the server inserts a brand new structure per call and cannot continue the previous one. Split points are chosen strictly by how much they change the rendered document — fully safe boundaries (blank lines, block starts that interrupt a paragraph) before boundaries that need repair (a table's rows now carry a re-emitted header and delimiter row; a fenced code block is closed and reopened with its original marker and info string) before boundaries that merely restructure (long paragraphs, list items) before a hard character cut. Within a tier the latest boundary in the window wins, since all chunks land in the same document. Every boundary that changes the rendered structure is reported in a new `degradations` field instead of being applied silently.
- **Fixed markdown chunking dropping a newline** — the previous splitter rebuilt block text from lines and lost one `\n` whenever the content's last line began a heading, table or code fence, so `"para\n# Title"` was written as `"para# Title"` and the heading stopped being a heading. Roughly one in five randomly generated documents was affected. The new splitter slices by offset and never rebuilds text, making content preservation structural.
- **Fixed oversized tables and code blocks being cut mid-cell and mid-fence** — the hard-split path never received the block type, so it cut at arbitrary character boundaries despite claiming to preserve table and code block integrity.
- **Fixed readback verification comparing against content the server never receives** — `doc +create` / `doc +update` verified the readback against the raw input, so any repaired boundary (and, previously, any paragraph split) failed verification on large documents. Verification now compares against the document the chunk plan says the server should hold.
- **Unified four markdown write paths onto one splitter** — `doc create` / `doc update`, `doc +create` / `doc +update` and `doc +checkpoint-update` now share `helpers.SplitMarkdownForAppend` and one limit constant (30000 runes), replacing two independent implementations plus one path that never chunked at all. `doc +checkpoint-update` accepts `@file` and stdin content, so oversized input was reachable there while the equivalent `doc +update` chunked. `doc +doc-append` takes `--text` from argv only and now rejects oversized input with a pointer to `doc +update` rather than sending one oversized call.
- **`doc update --index` now fails closed when the content requires chunking** — each chunk creates an unpredictable number of blocks, so the insertion point for later chunks is unknowable; the flag was previously accepted and silently ignored.
- **Chat automatic pagination controls** (#970) — adds bounded `--max-items` and cancellable `--page-delay` support to the core IM list shortcuts, with safe continuation metadata and truncation reporting.
- **Doc/drive/wiki routing descriptions** — clarifies the document-space container-vs-content boundary across the doc, drive, and wiki skill descriptions for more predictable first-round Agent selection, without changing CLI behavior.
- **Doc and Drive parameter aliases** — normalizes reviewed identifier, pagination, path, version, and role synonyms while blocking ambiguous values before dispatch.
- **International DingTalk region support** — adds `.io` login and MCP routing, pre-release endpoint overrides, and profile-aware gateway selection while preserving the existing `.com` flow.
- **Chat IM ID flags** (#954) — standardizes chat command entry points on `--conversation-id` for conversation IDs and `--message-id` for message IDs, so help, Schema, and Agent recommendations use the same canonical flags.
- **Legacy chat flag compatibility** (#954) — keeps older chat IM ID flags such as `--group`, `--id`, `--chat`, `--open-conversation-id`, `--msg-id`, and `--open-message-id` working as compatibility aliases where applicable, while hiding migrated aliases from recommended help and Schema surfaces.
- **Chat group bots target flag** (#954) — keeps `dws chat group bots` on the visible `--group` flag; this command does not register `--group-name`, and `--group` accepts either an openConversationId or a uniquely resolved group name.
- **Chat card update evidence** — distinguishes an accepted update request from an independently verified visible update, preserving the real `bizId` and warning callers not to repeat an unverified write.
- **Chat command guidance** — splits message and group references by task and explains that `--from` is ambiguous between sender and time-range intent.
- **Robot group reference replies** (#928) — `chat message send-by-bot` supports paired `--reply` and `--ref-sender` flags for Markdown replies that quote an existing group message.
- **Document write verification** (#960) — avoids false partial-success results when normalized Markdown, paginated blocks, inline images, or version reverts are confirmed by server readback. Document reverts and media inserts now require explicit readback evidence and report partial success when the server cannot prove the requested result.
- **Doc/drive description scope** — restates the `dingtalk-doc` description as document-entity-and-content operations with an explicit exclusion list, and narrows `dingtalk-drive` to file-level management of DingTalk documents, so first-round Agent selection separates content work from file management without changing CLI behavior.
- **Sheet SourceRange dropdowns** — supports range-backed dropdowns across direct, cell, and batch write paths, with structured readback for valid and invalid references. Batch `set-dropdown` now rejects unsupported top-level `colors` / `source-colors`; Inline colors belong in `options[].color`, while SourceRange color writes remain unsupported.
- **Sheet read completion metadata** — documents and preserves returned ranges, truncation reasons, and partial-read status for large range and CSV reads.
- **Windows event bus lifecycle** — start event consumers without unsupported inherited file descriptors, stop buses through local IPC with a termination fallback, and preserve subscription cleanup when startup fails.
- **Chat group roles** (#1058) — exposes the single-value `--role-id` flag for assigning one custom group role while preserving hidden `--role-ids` compatibility.
- **Chat user mentions** — preserves literal `<@openDingTalkId>` tokens in current-user Markdown messages and rejects mismatches between message-body mentions and mention flags before sending.
- **Chat direct media** — uses the IM upload target field for current-user direct file, audio, and video uploads, then uses the Chat receiver field for final message delivery.
- **CLI compatibility governance** — adds a reviewed two-stage path for hiding retained legacy commands or optional `NoOpt=true` boolean flags from Help and Schema when their activated capability moves to a dedicated command, with legacy-leaf, complete parameter/constant mapping, durable runtime constant evidence, protected framework bridges, dry-run preservation, parameter-collision, and fail-closed required-parameter checks.
- **OA admin approval query** — `oa approval list-by-admin` queries approval instances of a template with admin scope, with simple flags and an advanced `--request` mode; `startTime`/`endTime` use `yyyy-MM-dd HH:mm:ss` strings per the 2026-08 MCP contract update (ISO-8601 flag inputs auto-convert), and pageSize/time format are validated client-side with localized errors.
- **Chat personal emotions** — adds `chat emotion list`, `chat emotion send`, and `chat emotion favorite` for current-user personal favorite emotion listing, sending, and favoriting.
- **Minutes, DingTalk tasks, and Wiki parameter aliases** — adds reviewed parameter-name normalization, ambiguity guards, and end-to-end payload coverage for the three products.
- **Calendar empty windows** (#1074) — returns a legitimate empty result when the service emits its exact exhausted empty-event sentinel.
- **Task update verification** (#1074) — compares due-time readback as exact milliseconds so committed updates are no longer reported as failures.
- **Comment reaction validation** (#1074) — narrows accepted reaction input to reviewed DingTalk emoji names and rejects Unicode emoji and unsupported names such as `like` and `heart` before the RPC.
- **Stable release sealing** — directly preparing a stable release now renders and archives release fragments merged after its beta baseline, avoiding a forced extra beta solely to consume pending notes.
- **Whiteboard shortcuts** (#1082) — adds strict query and confirmed update workflows with stable-target receipts and exact readback verification.
- **Sheet shortcut hardening** (#1082) — makes worksheet listing and cell-range reads fail closed on malformed, ambiguous, or truncated responses, publishes a closed reviewed output shape, and preserves non-executing `--dry-run` previews for range reads.
- **AiSearch and Contact shortcuts** (#1083) — adds strict people search and reviewed unified results; people results must use the live-reviewed `person` source, and exact mobile lookups normalize accepted formatting before calling the dedicated mobile interface. Agent/public discovery keeps `contact +list-roles`, `contact +list-roster-fields`, `contact +get-roster`, and incomplete Live routes unavailable rather than publishing ambiguous results, while the historical Contact CLI commands retain legacy MCP execution and real error propagation. The legacy role-list projection preserves the service's reviewed null placeholder without exposing that ambiguous row through Agent Result contracts.
- **AITable datasource shortcuts** — adds 7 shortcuts for datasource sync management (`+datasource-create`, `+datasource-update`, `+datasource-sync`, `+datasource-sync-status`, `+datasource-get-config`, `+datasource-list-sources`, `+datasource-get-fields`) and updates the `dingtalk-aitable` skill with routing rules and a new `aitable-datasource.md` reference guide.
- **Legacy global slot recovery** — recovers a rejected identity refresh from the legacy global keychain slot when the organization mirror is absent, with strict corp/user matching so blank-user legacy tokens only recover for single-account organizations.
- **OA approval attachment upload** — `dws oa approval attachment upload --file <path>` uploads a local file as an approval attachment in one command: it initializes the upload credential (MCP `oa/init_attachment_upload_info`), HTTP PUTs the file to OSS, then commits it (MCP `oa/commit_attachment_upload_info`). `--file-name` defaults to the file's base name and `--md5` is auto-computed when omitted.
- **Sheet floating images** — supports creating or replacing a floating image directly from a local file with `create-float-image --file` and `update-float-image --file`, while retaining the existing `--src` workflow.
- **Sheet revision changesets** — adds read-only commands for querying the current workbook revision and reviewing Agent-readable changes between revisions, with guidance for distinguishing revisions from saved history versions and safely selecting rollback targets.
- **Education and college vendor extensions removed** — removes `dws edu-contact`, `dws edu-group`, `dws edu-app`, `dws edu-familygroup`, and `dws college-contact` from the CLI, Schema, bundled Skills, and open-edition MCP endpoint registry. Future DWS packages no longer expose these five command surfaces.
- **Reviewer Router merge recovery** — retries exact App-owned merge intents through a SHA-bound synchronous merge after GitHub has enforced approval and nine GitHub Actions source-bound required checks.
if test "${{ needs.dispatch-contract.outputs.mode }}" = plan_release; then
echo
echo "Plan only: no tag or package was created. Add the exact \`CHANGELOG.md\` section, merge it to main, then run publish."
echo "Plan only: no tag or package was created. Render pending \`.changes/*.md\` fragments into the exact \`CHANGELOG.md\` section, merge the release-seal PR to main, then run publish."
- Today (non-leaf): owning Cobra command → complete `corecmd.GroupPolicy{Mode, Positionals, Recovery}` → `corecmd.ApplyGroupPolicy`; the final assembled-tree gate rejects undeclared groups and stale group declarations on leaves
- **Declare = final Schema source**: `Flags` / `Constraints` / `Safety` / `ConstParams` / `Contract` (`corecmd.ContractDecl`; nested fields are `contract.*`)
- Naming: `ContractDecl` is the authoring leaf declaration. "Schema" means Catalog / `ToolSpec` delivery — do not reintroduce `SchemaDecl`.
-`Safety` uses `contract.SafetySpec` (`internal/corecmd/contract` only — no `cli.*` type alias). Its `confirmation` drives the runtime gate; `effect` / `risk` / `idempotency` are published unchanged. When `Contract` is set, convert once via `contractfinal.RegisterRuntimeContractFinal` (all callers — `corecmd.New` registers internally); assembly **pass-throughs** Final.
@@ -60,6 +61,7 @@ Schema contract) keep separate authorities — do not merge them with
- **Tier2** — `DeclareLeafMetadata` (helpers migration; **Shortcut may also use this path — acceptable**)
- **Tier3** — bare Cobra (should shrink over time; reviewed exclusions where needed)
- Long-term outlook only: broader mcpbind / fewer hand-written `Execute` bodies. **Not** a current hard requirement to delete `Shortcut.Execute` or force mcpbind.
- Group policy is separate from the leaf tiers: `corecmd.Spec` remains leaf-only. `ApplyGroupPolicy` must not infer or enable `TraverseChildren`; parent local-flag inheritance remains an explicit owning-command surface.
- Description declare vs delivery: construction requires `ContractDecl.Description` (evidence). Catalog delivery prefers Cobra Long → provenance `cobra_help`; without Long, declared text → `contract_final`. Title: declared first, then Short, then MCP. Do **not** read this as "declare = wire final" or dual authority.
- **Execute** = hooks (`Validate` / `Call` / `RunE` / `PostMount`) — not a second surface authority
- Declaration path has **no reviewed parallel fields**; migration-only `runtime_gate` annotate until `Safety` is declared
@@ -302,9 +304,8 @@ on the leaf:
```bash
dws auth status # token_valid should be true
dws cache refresh # deprecated no-op: prints a retirement notice (discovery cache is gone; refreshes nothing)
| `data_schema` | yes | One recursive JSON Schema **object** describing only the runtime envelope's `data` value. Every named `properties` child must have a non-empty `description`. It must not duplicate `ok`, `outcome`, `error`, or `meta`. |
| `sensitive_paths` | no | Unique safe dot paths relative to `data`; renderers/redaction consumers must not treat them as shell/JQ expressions. |
Optional members are omitted, never emitted as `null`. A leaf without a
reviewed Result omits the entire `result` key. Compact must preserve the same
normalized Result value as the full leaf; it must not summarize, infer, rename,
or independently rebuild any Result field. Product/group summaries do not
aggregate child Result objects.
`pagination` is a sibling of `result`, not a child. It declares the canonical
CLI cursor parameter and the fixed framework paths under `meta.pagination`.
Product response fields used to derive that metadata remain mapper internals;
they are not part of `result.data_schema`. Do not execute a second request to
derive pagination metadata.
Invalid result declarations fail closed during normalization: unknown or
duplicate outcomes, a non-object/multiple `data_schema`, unsafe or duplicate
sensitive paths, unsupported pagination kinds, attempts to override framework
meta paths, and an invalid cursor parameter must be rejected rather than
silently removed.
Full-leaf wire round trips must
preserve the normalized Result exactly. Do not commit generated Schema JSON as
evidence; tests construct contracts in Go and runtime/CI assemble the Catalog
from declarations.
### Performance model and rules
- Catalog construction is declaration-driven and cached through the existing
lazy `sync.Once` delivery path. Do not reassemble or reopen annotations per
command invocation, per leaf lookup, or per renderer.
- Normalizing one Result declaration is linear in the size of that declaration.
Full `schema --all` is linear in tools + parameters + Result schema bytes and
is an audit/compatibility export, not the normal Agent discovery path.
Overview → compact product/group → compact leaf remains the normal route;
only the final leaf carries its Result declaration.
- Constructing a `CommandResult` defensively clones result data and validates
invariants; rendering is buffer-first and then writes once. Both CPU cost and
transient memory are O(payload size), with roughly one additional in-memory
rendered copy. This buys immutability and prevents partial JSON leakage, but
it is not free.
- Large list/search commands must use bounded pages and publish continuation
facts. The current emitter buffers one command result/page before publishing;
pagination is the memory bound. Continuous event streams are a separate,
command-specific protocol and are not described by `ResultSpec`.
- A `dual_validate` command must execute the business request exactly once,
validate a shadow unified result, and preserve legacy bytes. Never obtain
validation by issuing a second network or write request.
- Filters and alternate formats are render-time work over the same in-memory
result. They must not rerun the business operation or rebuild Schema.
- Performance changes must preserve the one-result, buffer-first, fail-closed,
and atomic `--output` guarantees. Do not trade correctness for a microbenchmark
improvement. For a material hot-path change, benchmark representative small
and page-sized payloads and report allocations/bytes as well as latency.
## Current Schema boundaries
-`schema list` remains a progressive overview. `schema --all` is the stable
@@ -471,13 +602,16 @@ path; a generator unit test or JSON count alone is insufficient.
`parameters` object for commands without flags. Keep it suitable for the #602
compatibility baseline and fail rather than silently emitting a partial
export.
-`schema --all` is not normal command discovery. Use overview -> product/group
-> leaf for routine Agent work. `--compact` is supported for context-saving
projections, but a compact full export is not a complete compatibility
baseline.
-`schema --all` is not normal command discovery. Use overview -> compact
product/group -> compact leaf for routine Agent work. `--compact` is the
reviewed positive-field allowlist for Agent context: new full/audit fields
must not appear there until explicitly reviewed. A compact full export is not
a complete compatibility baseline.
-`dws <path> --help` defines whether Cobra exposes a path and which flags the
executable accepts. A leaf Schema defines Agent selection, parameter mapping
and constraints, and safety/confirmation semantics. A conflict is contract
drift, not permission to guess.
executable accepts. A compact leaf defines Agent selection, CLI parameters,
constraints, safety/confirmation semantics, and any reviewed `result`
contract. Full leaf fields such as `property`, `interface_ref`, and
provenance are audit facts. A conflict is contract drift, not permission to
guess.
- Schema and Help describe commands; neither returns DingTalk business data.
After discovery, execute the real read/search/list command to obtain data.
- **Switch later**: `dws skill setup --mode mono` (or `--mode multi`) — review the listed paths and confirm interactively.
</details>
@@ -210,7 +210,7 @@ The verifier uses isolated directories and does not replace the `dws` on the cur
The upgrade process follows a two-phase atomic flow to ensure consistency:
1. **Prepare** — downloads the platform-specific binary and skill packages to a temporary directory, verifies SHA256 checksums, and extracts/validates all files. If any step fails, the upgrade aborts without modifying the existing installation.
2. **Apply** — only after all preparations succeed, the binary is replaced and skill packages are installed to all detected agent directories (`~/.agents/skills/dws`, `~/.claude/skills/dws`, `~/.cursor/skills/dws`, etc.).
2. **Apply** — only after all preparations succeed, the binary is replaced and skills are flattened into the canonical `~/.agents/skills` root. Agents classified by the pinned compatibility registry as supporting the universal root read it directly; other detected Agents receive links to the canonical copy, with a direct-copy fallback when links are unavailable. Older DWS-managed agent-specific copies are backed up and retired so the same Skill is not discovered twice.
A backup of the current version is automatically created before each upgrade. Use `dws upgrade --rollback` to restore the previous version if needed.
Use Cobra help and Schema for different parts of the command contract:
- `dws <path> --help` is the source of truth for whether a command exists and which flags the binary accepts.
- `dws schema "<path>"` is the Agent contract for command selection, parameter mappings and constraints, risk, and confirmation semantics.
- `dws schema "<path>" --compact` is the normative Agent view for command selection, CLI parameters and constraints, risk, and confirmation; use a full leaf with a narrow `--jq` projection for mapping or provenance audits.
- If Help and Schema disagree, treat it as contract drift: pass only flags accepted by Cobra and use the more conservative safety semantics.
- Schema describes commands; it does not read or search DingTalk business data. Execute the real product command after discovery.
@@ -380,14 +380,14 @@ Use Cobra help and Schema for different parts of the command contract:
dws aitable record query --help
# Discover within a product, then inspect the selected leaf contract
dws schema aitable
dws schema "aitable record query"
dws schema aitable --compact
dws schema "aitable record query" --compact
# Execute the real business query
dws aitable record query --base-id BASE_ID --table-id TABLE_ID --limit 10
```
`dws schema --all` exports the complete contract for tooling, CI, audits, and compatibility baselines. Agents should prefer product/group discovery followed by a leaf query to avoid loading the full Catalog into context.
`dws schema --all` exports the complete contract for tooling, CI, audits, and compatibility baselines. Agents should query progressively with `--compact`; its positive field allowlist prevents new full/audit fields from silently expanding Agent context.
### Agent Skills
@@ -405,7 +405,7 @@ After installing, AI tools like Claude Code / Cursor can operate DingTalk direct
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-skills.sh | sh
```
> `install.sh` installs under`$HOME/.agents/skills/`(global; multi layout is per-product siblings, mono is the `dws/` subdirectory); `install-skills.sh` installs under `./.agents/skills/` (current project).
> Installers use`$HOME/.agents/skills/`as the canonical global store, following the universal `.agents/skills` convention. Agents classified by the pinned compatibility registry as universal read that root directly; detected non-universal Agents receive links to it (or copies when links are unavailable). Multi layout is per-product siblings, while mono uses the `dws/` subdirectory.
>
> China users: prefix `DWS_GITEE_REPO` to use the Gitee mirror — see [China mirror](#china-mirror).
| `--yes` | — | Scripting-only: skip the confirmation prompt. Removals are still backed up to `~/.dws/skill-backups/` first |
> The setup command can remove the opposite-mode layout (`dws/` for multi, DWS-managed multi Skills for mono) and stale managed Skills not in the bundle. DWS records ownership, installer version, source, and content digest centrally in `~/.dws/skills-state.json` (or `$DWS_CONFIG_DIR/skills-state.json`). Exact official names shipped before the centralized state remain a frozen migration list. A `dingtalk-*` prefix alone never authorizes cleanup, so other same-prefix market/user Skills are preserved. Every removal is previewed before confirmation and preserved under `~/.dws/skill-backups/<timestamp>/`; a directory that cannot be backed up is never removed. In a non-interactive shell, first run `--dry-run` and inspect its output; only then may the caller explicitly choose the scripting-only confirmation bypass.
After a multi setup or upgrade, DWS stores the official bundle snapshot and centralized ownership metadata in `~/.dws/skills-state.json` (or `$DWS_CONFIG_DIR/skills-state.json`). Every upgrade installs and overwrites the complete bundled Skill set from that release. Deleting or excluding a bundled Skill is not sticky: the next upgrade restores it. `dws upgrade --force` additionally allows reinstalling the current CLI version when no newer version is available.
Env vars: `DWS_SKILL_MODE=mono|multi` (also honored by `install.sh` / `install.ps1`), `DWS_SKILL_SOURCE=<path>`.
<summary><strong>Personal Event Subscription</strong> — real-time DingTalk messages for event-driven agents</summary>
`dws event consume` subscribes as the currently logged-in user over a managed Stream WebSocket and emits each event as one NDJSON line on stdout. The public catalog covers scoped and all one-to-one/group messages, specified senders, read/recall/reaction events, and group title/disband lifecycle events.
`dws event consume` subscribes as the currently logged-in user over a managed Stream WebSocket and emits each event as one NDJSON line on stdout. The public catalog covers scoped and all one-to-one/group messages, specified senders, read/recall/reaction events, group lifecycle events, and seven OA approval task/instance events.
The default `ndjson`, `json`, and `pretty` output preserves the transport envelope (`type`, `event_type`, string `data`, and `headers`) for existing scripts; `compact` retains its existing processor. Add `--flatten` to emit the stable top-level business fields used by Agent workflows. `--format` controls JSON serialization; `--flatten` controls the data structure and cannot be combined with `-f raw` or `--debug-raw-events`.
@@ -484,28 +492,33 @@ For an event-focused installation, use the official convenience installer:
```bash
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-event.sh | sh
# Or install the standalone multi skill from an existing dws installation
dws skill setup --mode multi -s event
```
```bash
# Inspect the public personal event catalog and schema
dws schema --all # full export for CI/audit/baselines
```
@@ -719,7 +738,7 @@ See [`docs/robot-quickstart.md`](./docs/robot-quickstart.md) for the full 4-step
<summary>Coming soon</summary>
- `conference` (video meetings)
- Multi-skill mode (default) — per-product skills under `skills/multi/`; installs and upgrades default to it, `dws skill setup --mode mono` switches back
- Multi-skill mode (default) — per-product skills under `skills/multi/`; installs and upgrades default to it, `dws skill setup --mode mono` switches back after interactive confirmation
</details>
@@ -768,6 +787,7 @@ See [`docs/robot-quickstart.md`](./docs/robot-quickstart.md) for the full 4-step
## Reference & Docs
- [International DingTalk (`.io`) guide](./docs/international-region-guide.md) — international login, domestic/international profile switching, isolated testing, and troubleshooting
- [Command Index](./docs/command-index.md) — every runtime command with description and when-to-use guidance
do not suppress `pull_request_target`, subject to GitHub's separate
security-sensitive branch-name restriction described above. The low-trust
trigger is forbidden from writing the default-branch cache directly. Before
dispatching, it gives Actions event delivery one minute to expose a run from
the exact protected `.github/workflows/ci.yml` workflow and exits if that normal producer already
owns the SHA, avoiding a duplicate full-suite run. A successful CI producer
must hard-verify its exact cache key. If that run instead completes with any
non-success conclusion, a separate base-owned `workflow_run` dispatcher binds
the exact workflow ID/path, run ID/attempt, conclusion, repository, branch, and
head SHA before requesting repair. `workflow_run` also has read-only
default-branch cache access, so both dispatchers use the reviewed
`repository_dispatch` exception. The dispatched default-branch producer
revalidates the corresponding merged-PR or failed-CI identity before checkout,
restores only the exact target key, recomputes the complete profile on a miss,
and verifies `cache-hit=true` after saving. An hourly schedule refreshes the
event-time `main` SHA after direct break-glass pushes or cache eviction;
`workflow_dispatch` provides the same current-main repair on demand. The
dedicated App identity remains mandatory because events created by the built-in
`GITHUB_TOKEN` can suppress both the main push and the closed-PR event.
Keep `HOMEBREW_PR_TOKEN` repository-scoped with `Contents: write` and
`Pull requests: write` (the latter remains necessary for withdrawal rollback),
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
chichuan