diff --git a/.gitignore b/.gitignore
index 8c98ebf1..6e5a1506 100644
--- a/.gitignore
+++ b/.gitignore
@@ -42,5 +42,8 @@ dws.zip
 # 功能测试运行产物
 results.jsonl
 test/dev_functional/results.jsonl
+/auto-test/
+/eval-runs/
+/.qoder/
 .vercel
 .env*
diff --git a/auto-test/cli_to_mcp/README.md b/auto-test/cli_to_mcp/README.md
deleted file mode 100644
index b2bdd695..00000000
--- a/auto-test/cli_to_mcp/README.md
+++ /dev/null
@@ -1,16 +0,0 @@
-# cli_to_mcp smoke tests
-
-This directory contains lightweight command-to-tool contract tests for hardcoded
-DWS commands synced from `dws-wukong`.
-
-The tests do not call live DingTalk APIs. They exercise command help, validation,
-and `--dry-run` output so command paths and MCP argument mappings stay stable.
-
-Run with an already built binary:
-
-```bash
-DWS_BIN=/path/to/dws pytest auto-test/cli_to_mcp/testcases
-```
-
-If `DWS_BIN` is not set, the runner falls back to `go run ./cmd` from the repo
-root.
diff --git a/auto-test/cli_to_mcp/testcases/agoal/test_01_agoal_commands.py b/auto-test/cli_to_mcp/testcases/agoal/test_01_agoal_commands.py
deleted file mode 100644
index bc40a452..00000000
--- a/auto-test/cli_to_mcp/testcases/agoal/test_01_agoal_commands.py
+++ /dev/null
@@ -1,193 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_agoal_strategy_and_contract_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "strategy",
-            "list",
-            "--scope-type",
-            "PERSONAL",
-            "--scope-id",
-            "user123",
-            "--request-id",
-            "req-1",
-            "--dry-run",
-        )
-    )
-    assert "list_strategy_decodings" in output
-    assert dry_run_args(output) == {
-        "scopeType": "PERSONAL",
-        "openId": "user123",
-        "requestId": "req-1",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "strategy",
-            "update",
-            "--profile-id",
-            "profile123",
-            "--content",
-            '[{"id":"e1","title":{"title":"new"}}]',
-            "--dry-run",
-        )
-    )
-    assert "update_strategy_decoding" in output
-    assert dry_run_args(output) == {
-        "profileId": "profile123",
-        "content": [{"id": "e1", "title": {"title": "new"}}],
-    }
-
-    output = assert_ok(dws.run_raw("agoal", "contract", "fields", "--dry-run"))
-    assert "list_op_contract_fields" in output
-    assert dry_run_args(output) == {}
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "contract",
-            "update",
-            "--contract-id",
-            "contract123",
-            "--dimensions",
-            '[{"id":"dim1","title":"metric"}]',
-            "--audit-config",
-            '{"needAudit":true}',
-            "--objective-template",
-            '{"id":"tpl1"}',
-            "--dry-run",
-        )
-    )
-    assert "update_op_contract" in output
-    assert dry_run_args(output) == {
-        "contractId": "contract123",
-        "dimensions": [{"id": "dim1", "title": "metric"}],
-        "auditConfig": '{"needAudit":true}',
-        "objectiveTemplate": '{"id":"tpl1"}',
-    }
-
-
-def test_agoal_scorecard_user_report_template_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "scorecard",
-            "detail",
-            "--selected-time",
-            "2026-01-01T00:00:00+08:00",
-            "--dept-id",
-            "dept123",
-            "--dry-run",
-        )
-    )
-    assert "get_score_card_detail" in output
-    args = dry_run_args(output)
-    assert args["deptId"] == "dept123"
-    assert args["selectedTime"] == 1767196800000
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "scorecard",
-            "update",
-            "--dept-id",
-            "dept123",
-            "--selected-time",
-            "2026-01-01",
-            "--id",
-            "sc123",
-            "--tracking-period-type",
-            "MONTHLY",
-            "--content",
-            '[{"id":"dim1","items":[]}]',
-            "--dry-run",
-        )
-    )
-    assert "update_score_card" in output
-    args = dry_run_args(output)
-    assert args["selectedTime"] == 1767196800000
-    assert args["content"] == [{"id": "dim1", "items": []}]
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "user",
-            "objectives",
-            "--user-id",
-            "user123",
-            "--rule-id",
-            "rule123",
-            "--period-ids",
-            "p1,p2",
-            "--dry-run",
-        )
-    )
-    assert "list_user_objectives" in output
-    assert dry_run_args(output) == {
-        "dingUserId": "user123",
-        "objectiveRuleId": "rule123",
-        "periodIds": ["p1", "p2"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "report",
-            "submit-detail",
-            "--template-id",
-            "tpl123",
-            "--submit-state",
-            "LATE",
-            "--query-date",
-            "2026-06-18T00:00:00+08:00",
-            "--page",
-            "1",
-            "--page-size",
-            "20",
-            "--keyword",
-            "alice",
-            "--dry-run",
-        )
-    )
-    assert "get_submit_detail" in output
-    assert dry_run_args(output) == {
-        "templateId": "tpl123",
-        "submitState": "LATE",
-        "queryDate": "2026-06-18",
-        "page": 1,
-        "pageSize": 20,
-        "keyword": "alice",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "agoal",
-            "obj-template",
-            "create-or-update",
-            "--title",
-            "tpl",
-            "--dimensions",
-            '[{"title":"dim"}]',
-            "--objective-weight",
-            "--dimension-weight",
-            "--compute-by-weight",
-            "--dry-run",
-        )
-    )
-    assert "create_or_update_obj_template" in output
-    assert dry_run_args(output) == {
-        "title": "tpl",
-        "dimensions": '[{"title":"dim"}]',
-        "objectiveWeight": True,
-        "dimensionWeight": True,
-        "computeByWeight": True,
-    }
diff --git a/auto-test/cli_to_mcp/testcases/chat/test_29_group_notice_and_new_commands.py b/auto-test/cli_to_mcp/testcases/chat/test_29_group_notice_and_new_commands.py
deleted file mode 100644
index 2fda8c65..00000000
--- a/auto-test/cli_to_mcp/testcases/chat/test_29_group_notice_and_new_commands.py
+++ /dev/null
@@ -1,163 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_group_notice_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "create",
-            "--group",
-            "cid123",
-            "--content",
-            "maintenance tonight",
-            "--sticky",
-            "--send-ding",
-            "--dry-run",
-        )
-    )
-    assert "create_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "content": "maintenance tonight",
-        "sticky": True,
-        "sendDing": True,
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "edit",
-            "--group",
-            "cid123",
-            "--notice-id",
-            "notice123",
-            "--content",
-            "updated",
-            "--dry-run",
-        )
-    )
-    assert "edit_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "dataId": "notice123",
-        "content": "updated",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "get",
-            "--group",
-            "cid123",
-            "--notice-id",
-            "notice123",
-            "--dry-run",
-        )
-    )
-    assert "get_group_notice" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "dataId": "notice123",
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "notice",
-            "list",
-            "--group",
-            "cid123",
-            "--limit",
-            "20",
-            "--cursor",
-            "next",
-            "--scheduled",
-            "--dry-run",
-        )
-    )
-    assert "list_group_notices" in output
-    assert dry_run_args(output) == {
-        "openConversationId": "cid123",
-        "limit": 20,
-        "cursor": "next",
-        "scheduled": True,
-    }
-
-
-def test_chat_misc_new_commands_cli_to_mcp(dws):
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "group",
-            "share-invite",
-            "--source",
-            "sourceCid",
-            "--target",
-            "targetCid",
-            "--expires-seconds",
-            "3600",
-            "--uuid",
-            "uuid-1",
-            "--dry-run",
-        )
-    )
-    assert "share_group_invite_url" in output
-    assert dry_run_args(output) == {
-        "sourceOpenConversationId": "sourceCid",
-        "targetOpenConversationId": "targetCid",
-        "expiresSeconds": 3600,
-        "uuid": "uuid-1",
-    }
-
-    output = assert_ok(
-        dws.run_raw("chat", "text", "translate", "--query", "hello", "--to", "zh_CN", "--dry-run")
-    )
-    assert "translate" in output
-    assert dry_run_args(output) == {"query": "hello", "to": "zh_CN"}
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "category",
-            "create-smart",
-            "--name",
-            "priority",
-            "--keywords",
-            "alpha,beta",
-            "--members",
-            "uid1,uid2",
-            "--dry-run",
-        )
-    )
-    assert "create_smart_conv_category" in output
-    assert dry_run_args(output) == {
-        "title": "priority",
-        "keywords": ["alpha", "beta"],
-        "memberOpenDingTalkIds": ["uid1", "uid2"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "chat",
-            "message",
-            "list-emotion-replies",
-            "--msg-ids",
-            "msg1,msg2",
-            "--dry-run",
-        )
-    )
-    assert "list_message_emotion_replies" in output
-    assert dry_run_args(output) == {"openMessageIds": ["msg1", "msg2"]}
diff --git a/auto-test/cli_to_mcp/testcases/conftest.py b/auto-test/cli_to_mcp/testcases/conftest.py
deleted file mode 100644
index c1ce4468..00000000
--- a/auto-test/cli_to_mcp/testcases/conftest.py
+++ /dev/null
@@ -1,8 +0,0 @@
-import pytest
-
-from test_utils import DWSRunner
-
-
-@pytest.fixture(scope="session")
-def dws():
-    return DWSRunner()
diff --git a/auto-test/cli_to_mcp/testcases/doc/test_12_import.py b/auto-test/cli_to_mcp/testcases/doc/test_12_import.py
deleted file mode 100644
index 80143d0f..00000000
--- a/auto-test/cli_to_mcp/testcases/doc/test_12_import.py
+++ /dev/null
@@ -1,54 +0,0 @@
-from test_utils import combined_output
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_doc_import_help_and_validation(dws, tmp_path):
-    output = assert_ok(dws.run_raw("doc", "import", "--help"))
-    assert "dws doc import" in output
-    assert "--file string" in output
-    assert "--workspace string" in output
-    assert "--name string" in output
-
-    result = dws.run_raw("doc", "import", "--file", str(tmp_path / "missing.md"), "--dry-run")
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "cannot read file" in output
-
-    bad = tmp_path / "bad.exe"
-    bad.write_text("bad", encoding="utf-8")
-    result = dws.run_raw("doc", "import", "--file", str(bad), "--dry-run")
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "unsupported file format" in output
-
-
-def test_doc_import_dry_run(dws, tmp_path):
-    source = tmp_path / "sample.md"
-    source.write_text("# Sample\n\nhello\n", encoding="utf-8")
-
-    output = assert_ok(
-        dws.run_raw(
-            "doc",
-            "import",
-            "--file",
-            str(source),
-            "--name",
-            "Imported Sample",
-            "--workspace",
-            "workspace123",
-            "--dry-run",
-        )
-    )
-    assert "Imported Sample" in output
-    assert "sample.md" in output
-    assert "md" in output
-
-
-def test_doc_import_get_dry_run(dws):
-    output = assert_ok(dws.run_raw("doc", "import", "get", "--task-id", "task123", "--dry-run"))
-    assert "task123" in output
diff --git a/auto-test/cli_to_mcp/testcases/mail/test_11_mail_new_commands.py b/auto-test/cli_to_mcp/testcases/mail/test_11_mail_new_commands.py
deleted file mode 100644
index 1a6bd5d0..00000000
--- a/auto-test/cli_to_mcp/testcases/mail/test_11_mail_new_commands.py
+++ /dev/null
@@ -1,143 +0,0 @@
-import os
-
-from test_utils import combined_output, dry_run_args
-
-
-def mail_email() -> str:
-    return os.environ.get("DINGTALK_MAIL_EMAIL", "user@example.com")
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def assert_fails(result, expected: str):
-    output = combined_output(result)
-    assert result.returncode != 0, output
-    assert expected in output
-
-
-def test_mailbox_profile_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "mailbox", "profile", "--help"))
-    assert "dws mail mailbox profile" in output
-    assert "--email string" in output
-
-    assert_fails(dws.run_raw("mail", "mailbox", "profile"), "email")
-
-    output = assert_ok(
-        dws.run_raw("mail", "mailbox", "profile", "--email", mail_email(), "--dry-run")
-    )
-    assert "get_mailbox_profile" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-
-def test_message_batch_get_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "message", "batch-get", "--help"))
-    assert "dws mail message batch-get" in output
-    assert "--email string" in output
-    assert "--ids string" in output
-
-    assert_fails(
-        dws.run_raw("mail", "message", "batch-get", "--email", mail_email()),
-        "ids",
-    )
-
-    too_many_ids = ",".join(f"msg_{i:02d}" for i in range(21))
-    assert_fails(
-        dws.run_raw(
-            "mail",
-            "message",
-            "batch-get",
-            "--email",
-            mail_email(),
-            "--ids",
-            too_many_ids,
-            "--dry-run",
-        ),
-        "20",
-    )
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "message",
-            "batch-get",
-            "--email",
-            mail_email(),
-            "--ids",
-            "msg_001,msg_002",
-            "--dry-run",
-        )
-    )
-    assert "get_email_by_message_id" in output
-    assert "msg_001" in output
-    assert "msg_002" in output
-
-
-def test_sent_message_recall_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "sent-message", "recall", "--help"))
-    assert "dws mail sent-message recall" in output
-    assert "--subject string" in output
-    assert "--yes" in output
-
-    assert_fails(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall",
-            "--email",
-            mail_email(),
-            "--id",
-            "msg_001",
-            "--subject",
-            "subject",
-        ),
-        "--yes",
-    )
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall",
-            "--email",
-            mail_email(),
-            "--id",
-            "msg_001",
-            "--subject",
-            "subject",
-            "--yes",
-            "--dry-run",
-        )
-    )
-    assert "recall_sent_message" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "id": "msg_001",
-        "subject": "subject",
-    }
-
-
-def test_sent_message_recall_detail_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "sent-message", "recall-detail", "--help"))
-    assert "dws mail sent-message recall-detail" in output
-    assert "--email string" in output
-    assert "--id string" in output
-    assert "FINISHED" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "sent-message",
-            "recall-detail",
-            "--email",
-            mail_email(),
-            "--id",
-            "task_001",
-            "--dry-run",
-        )
-    )
-    assert "get_recall_detail" in output
-    assert dry_run_args(output) == {"email": mail_email(), "id": "task_001"}
diff --git a/auto-test/cli_to_mcp/testcases/mail/test_12_mail_allowlist_blocklist_autoreply.py b/auto-test/cli_to_mcp/testcases/mail/test_12_mail_allowlist_blocklist_autoreply.py
deleted file mode 100644
index 2046693c..00000000
--- a/auto-test/cli_to_mcp/testcases/mail/test_12_mail_allowlist_blocklist_autoreply.py
+++ /dev/null
@@ -1,133 +0,0 @@
-import os
-
-from test_utils import combined_output, dry_run_args
-
-
-def mail_email() -> str:
-    return os.environ.get("DINGTALK_MAIL_EMAIL", "user@example.com")
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_auto_reply_update_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "auto-reply", "update", "--help"))
-    for flag in ("--email string", "--enabled string", "--start string", "--end string", "--scope string", "--content string"):
-        assert flag in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "auto-reply",
-            "update",
-            "--email",
-            mail_email(),
-            "--enabled",
-            "true",
-            "--start",
-            "2026/07/01 09:00:00 +0800",
-            "--end",
-            "2026/07/07 18:00:00 +0800",
-            "--scope",
-            "all",
-            "--content",
-            "out of office",
-            "--dry-run",
-        )
-    )
-    assert "update_auto_reply" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "enabled": True,
-        "startTime": "2026/07/01 09:00:00 +0800",
-        "endTime": "2026/07/07 18:00:00 +0800",
-        "scope": "all",
-        "content": "out of office",
-    }
-
-
-def test_allow_list_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "allow-list", "list", "--email", mail_email(), "--dry-run"))
-    assert "list_mailbox_allowlist" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "allow-list",
-            "add",
-            "--email",
-            mail_email(),
-            "--entries",
-            "partner@example.com,@example.org",
-            "--dry-run",
-        )
-    )
-    assert "add_mailbox_allowlist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["partner@example.com", "@example.org"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "allow-list",
-            "remove",
-            "--email",
-            mail_email(),
-            "--entries",
-            "partner@example.com",
-            "--dry-run",
-        )
-    )
-    assert "remove_mailbox_allowlist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["partner@example.com"],
-    }
-
-
-def test_block_list_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("mail", "block-list", "list", "--email", mail_email(), "--dry-run"))
-    assert "list_mailbox_blocklist" in output
-    assert dry_run_args(output) == {"email": mail_email()}
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "block-list",
-            "add",
-            "--email",
-            mail_email(),
-            "--entries",
-            "spam@example.com,@junk.example",
-            "--dry-run",
-        )
-    )
-    assert "add_mailbox_blocklist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["spam@example.com", "@junk.example"],
-    }
-
-    output = assert_ok(
-        dws.run_raw(
-            "mail",
-            "block-list",
-            "remove",
-            "--email",
-            mail_email(),
-            "--entries",
-            "spam@example.com",
-            "--dry-run",
-        )
-    )
-    assert "remove_mailbox_blocklist" in output
-    assert dry_run_args(output) == {
-        "email": mail_email(),
-        "entries": ["spam@example.com"],
-    }
diff --git a/auto-test/cli_to_mcp/testcases/sheet/test_36_group_dimension.py b/auto-test/cli_to_mcp/testcases/sheet/test_36_group_dimension.py
deleted file mode 100644
index b2737490..00000000
--- a/auto-test/cli_to_mcp/testcases/sheet/test_36_group_dimension.py
+++ /dev/null
@@ -1,80 +0,0 @@
-from test_utils import combined_output, dry_run_args
-
-
-def assert_ok(result):
-    output = combined_output(result)
-    assert result.returncode == 0, output
-    return output
-
-
-def test_group_dimension_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("sheet", "group-dimension", "--help"))
-    assert "dws sheet group-dimension" in output
-    assert "--group-state string" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "sheet",
-            "group-dimension",
-            "--node",
-            "node123",
-            "--sheet-id",
-            "Sheet1",
-            "--range",
-            "3:7",
-            "--group-state",
-            "fold",
-            "--dry-run",
-        )
-    )
-    assert "group_dimension" in output
-    assert dry_run_args(output) == {
-        "nodeId": "node123",
-        "sheetId": "Sheet1",
-        "range": "3:7",
-        "groupState": "fold",
-    }
-
-
-def test_ungroup_dimension_cli_to_mcp(dws):
-    output = assert_ok(dws.run_raw("sheet", "ungroup-dimension", "--help"))
-    assert "dws sheet ungroup-dimension" in output
-
-    output = assert_ok(
-        dws.run_raw(
-            "sheet",
-            "ungroup-dimension",
-            "--node",
-            "node123",
-            "--sheet-id",
-            "Sheet1",
-            "--range",
-            "C:F",
-            "--dry-run",
-        )
-    )
-    assert "ungroup_dimension" in output
-    assert dry_run_args(output) == {
-        "nodeId": "node123",
-        "sheetId": "Sheet1",
-        "range": "C:F",
-    }
-
-
-def test_group_dimension_rejects_invalid_state(dws):
-    result = dws.run_raw(
-        "sheet",
-        "group-dimension",
-        "--node",
-        "node123",
-        "--sheet-id",
-        "Sheet1",
-        "--range",
-        "3:7",
-        "--group-state",
-        "invalid",
-        "--dry-run",
-    )
-    output = combined_output(result)
-    assert result.returncode != 0
-    assert "group-state" in output
diff --git a/auto-test/cli_to_mcp/testcases/test_utils/__init__.py b/auto-test/cli_to_mcp/testcases/test_utils/__init__.py
deleted file mode 100644
index 377beca2..00000000
--- a/auto-test/cli_to_mcp/testcases/test_utils/__init__.py
+++ /dev/null
@@ -1,58 +0,0 @@
-import json
-import os
-import re
-import shlex
-import subprocess
-from pathlib import Path
-
-
-def repo_root(start_file: str) -> Path:
-    current = Path(start_file).resolve()
-    for parent in [current, *current.parents]:
-        if (parent / "go.mod").exists():
-            return parent
-    raise RuntimeError(f"cannot locate repo root from {start_file}")
-
-
-def resolve_dws_cmd(start_file: str) -> list[str]:
-    root = repo_root(start_file)
-    if env_bin := os.environ.get("DWS_BIN"):
-        return shlex.split(env_bin)
-
-    for rel in ("dws", "build/dws", "bin/dws", "dingtalk-workspace-cli"):
-        candidate = root / rel
-        if candidate.exists() and os.access(candidate, os.X_OK):
-            return [str(candidate)]
-
-    return ["go", "run", "./cmd"]
-
-
-def combined_output(result: subprocess.CompletedProcess) -> str:
-    return (result.stdout or "") + (result.stderr or "")
-
-
-def dry_run_args(output: str) -> dict:
-    match = re.search(r"Arguments:\s*(\{.*\})", output, re.S)
-    assert match, f"dry-run output does not contain Arguments JSON: {output}"
-    return json.loads(match.group(1))
-
-
-class DWSRunner:
-    def __init__(self):
-        self.root = repo_root(__file__)
-        self.cmd = resolve_dws_cmd(__file__)
-
-    def run_raw(self, *args: str, timeout: int = 45) -> subprocess.CompletedProcess:
-        return subprocess.run(
-            [*self.cmd, *args],
-            cwd=self.root,
-            text=True,
-            capture_output=True,
-            timeout=timeout,
-        )
-
-    def run(self, *args: str, timeout: int = 45):
-        result = self.run_raw(*args, timeout=timeout)
-        output = combined_output(result)
-        assert result.returncode == 0, output
-        return json.loads(result.stdout)
diff --git a/internal/helpers/chat.go b/internal/helpers/chat.go
index b98f08b2..d31690d7 100644
--- a/internal/helpers/chat.go
+++ b/internal/helpers/chat.go
@@ -15,6 +15,7 @@ import (
 	"strings"
 	"time"
 
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
 	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
 	"github.com/spf13/cobra"
 )
@@ -4927,17 +4928,26 @@ status 可选值:
   dws chat category create-smart --name "团队群" --members openDingTalkId1,openDingTalkId2
   dws chat category create-smart --name "重点群" --keywords "重点" --members openDingTalkId1`,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			if err := validateRequiredFlags(cmd, "name"); err != nil {
-				return err
+			name := strings.TrimSpace(mustGetFlag(cmd, "name"))
+			if name == "" {
+				return apperrors.NewValidation("--name must not be blank")
 			}
 			toolArgs := map[string]any{
-				"title": mustGetFlag(cmd, "name"),
+				"categoryName": name,
 			}
-			if v, _ := cmd.Flags().GetString("keywords"); v != "" {
-				toolArgs["keywords"] = parseCSVValues(v)
+			if cmd.Flags().Changed("keywords") {
+				values := parseCSVValues(mustGetFlag(cmd, "keywords"))
+				if len(values) == 0 {
+					return apperrors.NewValidation("--keywords must contain at least one non-empty value")
+				}
+				toolArgs["groupNameKeywords"] = values
 			}
-			if v, _ := cmd.Flags().GetString("members"); v != "" {
-				toolArgs["memberOpenDingTalkIds"] = parseCSVValues(v)
+			if cmd.Flags().Changed("members") {
+				values := parseCSVValues(mustGetFlag(cmd, "members"))
+				if len(values) == 0 {
+					return apperrors.NewValidation("--members must contain at least one non-empty value")
+				}
+				toolArgs["memberOpenDingTalkIds"] = values
 			}
 			return callMCPToolOnServer("im", "create_smart_conv_category", toolArgs)
 		},
diff --git a/internal/helpers/chat_message_search_test.go b/internal/helpers/chat_message_search_test.go
new file mode 100644
index 00000000..f036ef71
--- /dev/null
+++ b/internal/helpers/chat_message_search_test.go
@@ -0,0 +1,129 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package helpers
+
+import (
+	"context"
+	"io"
+	"os"
+	"reflect"
+	"testing"
+	"time"
+
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+type chatMessageSearchCall struct {
+	productID string
+	toolName  string
+	args      map[string]any
+}
+
+type chatMessageSearchCaller struct {
+	calls []chatMessageSearchCall
+}
+
+func (c *chatMessageSearchCaller) CallTool(_ context.Context, productID, toolName string, args map[string]any) (*edition.ToolResult, error) {
+	c.calls = append(c.calls, chatMessageSearchCall{productID: productID, toolName: toolName, args: args})
+	return &edition.ToolResult{Content: []edition.ContentBlock{{Type: "text", Text: `{}`}}}, nil
+}
+
+func (*chatMessageSearchCaller) Format() string { return "json" }
+func (*chatMessageSearchCaller) DryRun() bool   { return false }
+func (*chatMessageSearchCaller) Fields() string { return "" }
+func (*chatMessageSearchCaller) JQ() string     { return "" }
+
+func TestChatMessageSearchUsesMCPContracts(t *testing.T) {
+	previousDeps := deps
+	previousArgs := os.Args
+	t.Cleanup(func() {
+		deps = previousDeps
+		os.Args = previousArgs
+	})
+
+	start := "2026-07-09T00:00:00+08:00"
+	end := "2026-07-11T00:00:00+08:00"
+	startTime, err := time.Parse(time.RFC3339, start)
+	if err != nil {
+		t.Fatalf("parse start time: %v", err)
+	}
+	endTime, err := time.Parse(time.RFC3339, end)
+	if err != nil {
+		t.Fatalf("parse end time: %v", err)
+	}
+
+	tests := []struct {
+		name        string
+		args        []string
+		productID   string
+		toolName    string
+		wantToolArg map[string]any
+	}{
+		{
+			name:      "keyword search",
+			args:      []string{"message", "search", "--query", "categoryName", "--group", "cid-1", "--start", start, "--end", end, "--limit", "100", "--cursor", "0"},
+			productID: "chat",
+			toolName:  "search_messages_by_keyword",
+			wantToolArg: map[string]any{
+				"keyword":            "categoryName",
+				"openConversationId": "cid-1",
+				"startTime":          startTime.UnixMilli(),
+				"endTime":            endTime.UnixMilli(),
+				"limit":              100,
+				"cursor":             "0",
+			},
+		},
+		{
+			name:      "advanced search",
+			args:      []string{"message", "search-advanced", "--query", "categoryName", "--conversation-ids", "cid-1,cid-2", "--start", start, "--end", end, "--limit", "100", "--cursor", "0"},
+			productID: "im",
+			toolName:  "search_messages",
+			wantToolArg: map[string]any{
+				"keyword":             "categoryName",
+				"openConversationIds": []string{"cid-1", "cid-2"},
+				"startTime":           startTime.UnixMilli(),
+				"endTime":             endTime.UnixMilli(),
+				"limit":               100,
+				"cursor":              "0",
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			caller := &chatMessageSearchCaller{}
+			InitDeps(caller)
+			deps.Out.w = io.Discard
+			os.Args = []string{"dws", "chat"}
+
+			cmd := newChatCommand()
+			cmd.SilenceErrors = true
+			cmd.SilenceUsage = true
+			cmd.SetArgs(tt.args)
+			if err := cmd.Execute(); err != nil {
+				t.Fatalf("chat search returned error: %v", err)
+			}
+			if len(caller.calls) != 1 {
+				t.Fatalf("tool call count = %d, want 1", len(caller.calls))
+			}
+			call := caller.calls[0]
+			if call.productID != tt.productID || call.toolName != tt.toolName {
+				t.Fatalf("tool call = %s/%s, want %s/%s", call.productID, call.toolName, tt.productID, tt.toolName)
+			}
+			if !reflect.DeepEqual(call.args, tt.wantToolArg) {
+				t.Fatalf("tool args = %#v, want %#v", call.args, tt.wantToolArg)
+			}
+		})
+	}
+}
diff --git a/internal/helpers/chat_smart_category_test.go b/internal/helpers/chat_smart_category_test.go
new file mode 100644
index 00000000..abb7d499
--- /dev/null
+++ b/internal/helpers/chat_smart_category_test.go
@@ -0,0 +1,187 @@
+// Copyright 2026 Alibaba Group
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package helpers
+
+import (
+	"context"
+	"io"
+	"reflect"
+	"strings"
+	"testing"
+
+	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
+	"github.com/DingTalk-Real-AI/dingtalk-workspace-cli/pkg/edition"
+)
+
+type chatSmartCategoryCall struct {
+	productID string
+	toolName  string
+	args      map[string]any
+}
+
+type chatSmartCategoryCaller struct {
+	calls []chatSmartCategoryCall
+}
+
+func (c *chatSmartCategoryCaller) CallTool(_ context.Context, productID, toolName string, args map[string]any) (*edition.ToolResult, error) {
+	c.calls = append(c.calls, chatSmartCategoryCall{
+		productID: productID,
+		toolName:  toolName,
+		args:      args,
+	})
+	return &edition.ToolResult{Content: []edition.ContentBlock{{Type: "text", Text: `{}`}}}, nil
+}
+
+func (*chatSmartCategoryCaller) Format() string { return "json" }
+func (*chatSmartCategoryCaller) DryRun() bool   { return false }
+func (*chatSmartCategoryCaller) Fields() string { return "" }
+func (*chatSmartCategoryCaller) JQ() string     { return "" }
+
+func TestChatCategoryCreateSmartUsesMCPParameterNames(t *testing.T) {
+	previousDeps := deps
+	t.Cleanup(func() { deps = previousDeps })
+
+	caller := &chatSmartCategoryCaller{}
+	InitDeps(caller)
+	deps.Out.w = io.Discard
+
+	cmd := newChatCommand()
+	cmd.SilenceErrors = true
+	cmd.SilenceUsage = true
+	cmd.SetArgs([]string{
+		"category", "create-smart",
+		"--name", "priority",
+		"--keywords", "alpha, ,beta,",
+		"--members", "open-id-1, ,open-id-2,",
+	})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("chat category create-smart returned error: %v", err)
+	}
+
+	if len(caller.calls) != 1 {
+		t.Fatalf("tool call count = %d, want 1", len(caller.calls))
+	}
+	call := caller.calls[0]
+	if call.productID != "im" || call.toolName != "create_smart_conv_category" {
+		t.Fatalf("tool call = %s/%s, want im/create_smart_conv_category", call.productID, call.toolName)
+	}
+	want := map[string]any{
+		"categoryName":          "priority",
+		"groupNameKeywords":     []string{"alpha", "beta"},
+		"memberOpenDingTalkIds": []string{"open-id-1", "open-id-2"},
+	}
+	if !reflect.DeepEqual(call.args, want) {
+		t.Fatalf("tool args = %#v, want %#v", call.args, want)
+	}
+}
+
+func TestChatCategoryCreateSmartRejectsBlankInputs(t *testing.T) {
+	tests := []struct {
+		name    string
+		args    []string
+		wantErr string
+	}{
+		{
+			name:    "empty name",
+			args:    []string{"--name", ""},
+			wantErr: "--name must not be blank",
+		},
+		{
+			name:    "blank name",
+			args:    []string{"--name", " \t "},
+			wantErr: "--name must not be blank",
+		},
+		{
+			name:    "empty keywords",
+			args:    []string{"--name", "priority", "--keywords", ""},
+			wantErr: "--keywords must contain at least one non-empty value",
+		},
+		{
+			name:    "blank keywords CSV",
+			args:    []string{"--name", "priority", "--keywords", " , , "},
+			wantErr: "--keywords must contain at least one non-empty value",
+		},
+		{
+			name:    "empty keywords JSON array",
+			args:    []string{"--name", "priority", "--keywords", "[]"},
+			wantErr: "--keywords must contain at least one non-empty value",
+		},
+		{
+			name:    "empty members",
+			args:    []string{"--name", "priority", "--members", ""},
+			wantErr: "--members must contain at least one non-empty value",
+		},
+		{
+			name:    "blank members CSV",
+			args:    []string{"--name", "priority", "--members", " , , "},
+			wantErr: "--members must contain at least one non-empty value",
+		},
+		{
+			name:    "blank members JSON array",
+			args:    []string{"--name", "priority", "--members", `["", " "]`},
+			wantErr: "--members must contain at least one non-empty value",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			previousDeps := deps
+			t.Cleanup(func() { deps = previousDeps })
+
+			caller := &chatSmartCategoryCaller{}
+			InitDeps(caller)
+			deps.Out.w = io.Discard
+
+			cmd := newChatCommand()
+			cmd.SilenceErrors = true
+			cmd.SilenceUsage = true
+			cmd.SetArgs(append([]string{"category", "create-smart"}, tt.args...))
+			err := cmd.Execute()
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("error = %v, want containing %q", err, tt.wantErr)
+			}
+			if got := apperrors.ExitCode(err); got != 3 {
+				t.Fatalf("exit code = %d, want validation code 3", got)
+			}
+			if len(caller.calls) != 0 {
+				t.Fatalf("tool call count = %d, want 0", len(caller.calls))
+			}
+		})
+	}
+}
+
+func TestChatCategoryCreateSmartAllowsOmittedRulesAndTrimsName(t *testing.T) {
+	previousDeps := deps
+	t.Cleanup(func() { deps = previousDeps })
+
+	caller := &chatSmartCategoryCaller{}
+	InitDeps(caller)
+	deps.Out.w = io.Discard
+
+	cmd := newChatCommand()
+	cmd.SilenceErrors = true
+	cmd.SilenceUsage = true
+	cmd.SetArgs([]string{"category", "create-smart", "--name", "  priority  "})
+	if err := cmd.Execute(); err != nil {
+		t.Fatalf("chat category create-smart returned error: %v", err)
+	}
+
+	if len(caller.calls) != 1 {
+		t.Fatalf("tool call count = %d, want 1", len(caller.calls))
+	}
+	want := map[string]any{"categoryName": "priority"}
+	if !reflect.DeepEqual(caller.calls[0].args, want) {
+		t.Fatalf("tool args = %#v, want %#v", caller.calls[0].args, want)
+	}
+}
diff --git a/internal/i18n/locales/en.json b/internal/i18n/locales/en.json
index 1b509b80..0e4e7f36 100644
--- a/internal/i18n/locales/en.json
+++ b/internal/i18n/locales/en.json
@@ -105,6 +105,7 @@
   "截止时间 ISO-8601 (如 2026-03-10T18:00:00+08:00)": "Due time ISO-8601 (e.g. 2026-03-10T18:00:00+08:00)",
   "执行者 userId 列表 (必填)": "Executor userId list (required)",
   "检查登录状态后重试": "Check login status and retry",
+  "检查网络、代理和 DNS 配置后重试原命令": "Check network, proxy, and DNS settings, then retry the original command",
   "检查服务连通性后重试；如持续失败，请确认 MCP 服务响应正常。": "Check service connectivity and retry; if persistent, verify MCP service is responding.",
   "检查服务连通性和协议版本后重试": "Check service connectivity and protocol version, then retry",
   "检查认证、权限和参数后重试原命令": "Check authentication, permissions, and parameters, then retry",
@@ -117,6 +118,7 @@
   "完成状态: true/false": "Done status: true/false",
   "完成状态: true=已完成, false=未完成 (必填)": "Done status: true=completed, false=not completed (required)",
   "确认删除? (yes/no): ": "Confirm delete? (yes/no): ",
+  "确认 MCP 服务可访问；若持续失败请稍后重试": "Verify the MCP service is reachable; if the failure persists, retry later",
   "等待用户授权...": "Waiting for user authorization...",
   "结果格式与客户端预期不一致；请检查服务协议变更或回退到最近可用版本。": "Result format does not match client expectations; check service protocol changes or rollback to latest working version.",
   "网络错误，继续重试...": "Network error, retrying...",
diff --git a/internal/i18n/locales/zh.json b/internal/i18n/locales/zh.json
index cd463211..d4e491db 100644
--- a/internal/i18n/locales/zh.json
+++ b/internal/i18n/locales/zh.json
@@ -105,6 +105,7 @@
   "截止时间 ISO-8601 (如 2026-03-10T18:00:00+08:00)": "截止时间 ISO-8601 (如 2026-03-10T18:00:00+08:00)",
   "执行者 userId 列表 (必填)": "执行者 userId 列表 (必填)",
   "检查登录状态后重试": "检查登录状态后重试",
+  "检查网络、代理和 DNS 配置后重试原命令": "检查网络、代理和 DNS 配置后重试原命令",
   "检查服务连通性后重试；如持续失败，请确认 MCP 服务响应正常。": "检查服务连通性后重试；如持续失败，请确认 MCP 服务响应正常。",
   "检查服务连通性和协议版本后重试": "检查服务连通性和协议版本后重试",
   "检查认证、权限和参数后重试原命令": "检查认证、权限和参数后重试原命令",
@@ -117,6 +118,7 @@
   "完成状态: true/false": "完成状态: true/false",
   "完成状态: true=已完成, false=未完成 (必填)": "完成状态: true=已完成, false=未完成 (必填)",
   "确认删除? (yes/no): ": "确认删除? (yes/no): ",
+  "确认 MCP 服务可访问；若持续失败请稍后重试": "确认 MCP 服务可访问；若持续失败请稍后重试",
   "等待用户授权...": "等待用户授权...",
   "结果格式与客户端预期不一致；请检查服务协议变更或回退到最近可用版本。": "结果格式与客户端预期不一致；请检查服务协议变更或回退到最近可用版本。",
   "网络错误，继续重试...": "网络错误，继续重试...",
diff --git a/internal/transport/client.go b/internal/transport/client.go
index ed516765..8b9a2d8a 100644
--- a/internal/transport/client.go
+++ b/internal/transport/client.go
@@ -416,7 +416,7 @@ func (c *Client) callJSONRPC(ctx context.Context, endpoint string, request reque
 	}
 	callStart := time.Now()
 
-	resp, err := c.doWithRetry(ctx, endpoint, body)
+	resp, err := c.doWithRetry(ctx, endpoint, body, request.Method)
 	if err != nil {
 		return err
 	}
@@ -493,15 +493,19 @@ func (c *Client) callJSONRPC(ctx context.Context, endpoint string, request reque
 	return nil
 }
 
-func (c *Client) doWithRetry(ctx context.Context, endpoint string, body []byte) (*http.Response, error) {
+func (c *Client) doWithRetry(ctx context.Context, endpoint string, body []byte, method string) (*http.Response, error) {
 	endpoint = sanitizeJSONRPCEndpoint(endpoint)
+	operation := strings.TrimSpace(method)
+	if operation == "" {
+		operation = "jsonrpc"
+	}
 	var lastErr error
 	for attempt := 0; attempt <= c.MaxRetries; attempt++ {
 		req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(body))
 		if err != nil {
 			return nil, apperrors.NewDiscovery(
 				"failed to create JSON-RPC request",
-				apperrors.WithOperation("jsonrpc"),
+				apperrors.WithOperation(operation),
 				apperrors.WithReason("request_build_failed"),
 				apperrors.WithHint(i18n.T("请检查服务 endpoint 是否为空或格式不合法。")),
 				apperrors.WithCause(&CallError{
@@ -559,37 +563,52 @@ func (c *Client) doWithRetry(ctx context.Context, endpoint string, body []byte)
 				statusForLog = resp.StatusCode
 			}
 			delay := c.retryDelayForAttempt(attempt, retryAfter)
-			logging.LogRetryAttempt(c.FileLogger, "jsonrpc", c.ExecutionId, attempt, c.MaxRetries, statusForLog, delay, lastErr)
+			logging.LogRetryAttempt(c.FileLogger, operation, c.ExecutionId, attempt, c.MaxRetries, statusForLog, delay, lastErr)
 			if err := c.sleepForRetry(ctx, delay); err != nil {
-				return nil, apperrors.NewDiscovery(
-					"request cancelled during retry",
-					apperrors.WithOperation("jsonrpc"),
+				opts := []apperrors.Option{
+					apperrors.WithOperation(operation),
 					apperrors.WithReason("request_cancelled"),
 					apperrors.WithHint(i18n.T("请求在重试过程中被取消；请检查调用侧超时设置。")),
 					apperrors.WithCause(&CallError{
 						Stage: CallStageRequest,
 						Cause: err,
 					}),
-				)
+				}
+				if operation == "tools/call" {
+					opts = append(opts, apperrors.WithActions(networkActions("")...))
+					return nil, apperrors.NewAPI("request cancelled during retry", opts...)
+				}
+				opts = append(opts, apperrors.WithActions(discoveryActions("")...))
+				return nil, apperrors.NewDiscovery("request cancelled during retry", opts...)
 			}
 		}
 	}
 	reason, hint := classifyRequestFailure(lastErr)
-	logging.LogErrorClassified(c.FileLogger, "jsonrpc", c.ExecutionId,
-		string(apperrors.CategoryDiscovery), reason, 0, 0,
+	category := apperrors.CategoryDiscovery
+	actions := discoveryActions("")
+	if operation == "tools/call" {
+		category = apperrors.CategoryAPI
+		actions = networkActions("")
+	}
+	logging.LogErrorClassified(c.FileLogger, operation, c.ExecutionId,
+		string(category), reason, 0, 0,
 		!isTimeoutError(lastErr), "")
-	return nil, apperrors.NewDiscovery(
-		fmt.Sprintf("request to %s failed: %v", RedactURL(endpoint), lastErr),
-		apperrors.WithOperation("jsonrpc"),
+	opts := []apperrors.Option{
+		apperrors.WithOperation(operation),
 		apperrors.WithReason(reason),
 		apperrors.WithRetryable(!isTimeoutError(lastErr)),
 		apperrors.WithHint(hint),
-		apperrors.WithActions(discoveryActions("")...),
+		apperrors.WithActions(actions...),
 		apperrors.WithCause(&CallError{
 			Stage: CallStageRequest,
 			Cause: lastErr,
 		}),
-	)
+	}
+	message := fmt.Sprintf("request to %s failed: %v", RedactURL(endpoint), lastErr)
+	if category == apperrors.CategoryAPI {
+		return nil, apperrors.NewAPI(message, opts...)
+	}
+	return nil, apperrors.NewDiscovery(message, opts...)
 }
 
 func sanitizeJSONRPCEndpoint(endpoint string) string {
@@ -1039,6 +1058,17 @@ func runtimeActions(snapshotPath string) []string {
 	return actions
 }
 
+func networkActions(snapshotPath string) []string {
+	actions := []string{
+		i18n.T("检查网络、代理和 DNS 配置后重试原命令"),
+		i18n.T("确认 MCP 服务可访问；若持续失败请稍后重试"),
+	}
+	if snapshotPath != "" {
+		actions = append(actions, fmt.Sprintf("dws recovery plan --snapshot %s", snapshotPath))
+	}
+	return actions
+}
+
 // validateCallArguments checks string values in tool call arguments for
 // control characters and dangerous Unicode at the transport boundary.
 func validateCallArguments(args map[string]any) error {
diff --git a/internal/transport/client_extra_test.go b/internal/transport/client_extra_test.go
index 05177cbd..d9a1b070 100644
--- a/internal/transport/client_extra_test.go
+++ b/internal/transport/client_extra_test.go
@@ -103,7 +103,7 @@ func TestDoWithRetryRedactsGatewayQueryInHeaderDebugLog(t *testing.T) {
 	})})
 	client.FileLogger = logger
 
-	resp, err := client.doWithRetry(context.Background(), "https://mcp-gw.dingtalk.com/server/demo?key=secret#frag", []byte(`{}`))
+	resp, err := client.doWithRetry(context.Background(), "https://mcp-gw.dingtalk.com/server/demo?key=secret#frag", []byte(`{}`), "initialize")
 	if err != nil {
 		t.Fatalf("doWithRetry() error = %v", err)
 	}
diff --git a/internal/transport/recovery_metadata_test.go b/internal/transport/recovery_metadata_test.go
index cacccd81..5dfcd7c3 100644
--- a/internal/transport/recovery_metadata_test.go
+++ b/internal/transport/recovery_metadata_test.go
@@ -4,7 +4,9 @@ import (
 	"context"
 	"errors"
 	"net/http"
+	"strings"
 	"testing"
+	"time"
 
 	apperrors "github.com/DingTalk-Real-AI/dingtalk-workspace-cli/internal/errors"
 )
@@ -53,7 +55,7 @@ func TestDoWithRetryRequestFailureIncludesCallMetadata(t *testing.T) {
 	})})
 	client.MaxRetries = 0
 
-	_, err := client.doWithRetry(context.Background(), "https://mcp.dingtalk.com/server", []byte(`{}`))
+	_, err := client.doWithRetry(context.Background(), "https://mcp.dingtalk.com/server", []byte(`{}`), "tools/call")
 	if err == nil {
 		t.Fatal("doWithRetry() error = nil, want request failure")
 	}
@@ -67,6 +69,125 @@ func TestDoWithRetryRequestFailureIncludesCallMetadata(t *testing.T) {
 	}
 }
 
+func TestCallToolRequestFailureUsesAPIClassification(t *testing.T) {
+	client := NewClient(&http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+		return nil, errors.New("dial tcp: connection refused")
+	})})
+	client.MaxRetries = 0
+
+	_, err := client.CallTool(context.Background(), "https://mcp.dingtalk.com/server", "search_messages", map[string]any{"keyword": "marker"})
+	if err == nil {
+		t.Fatal("CallTool() error = nil, want request failure")
+	}
+
+	var typed *apperrors.Error
+	if !errors.As(err, &typed) {
+		t.Fatalf("CallTool() error = %T, want *errors.Error", err)
+	}
+	if typed.Category != apperrors.CategoryAPI {
+		t.Fatalf("Category = %q, want %q", typed.Category, apperrors.CategoryAPI)
+	}
+	if typed.Operation != "tools/call" {
+		t.Fatalf("Operation = %q, want tools/call", typed.Operation)
+	}
+	if typed.Reason != "connection_refused" {
+		t.Fatalf("Reason = %q, want connection_refused", typed.Reason)
+	}
+	actions := strings.Join(typed.Actions, "\n")
+	if strings.Contains(actions, "internal/syncdata") || strings.Contains(actions, "sync-oss") {
+		t.Fatalf("runtime network failure contains discovery-only actions: %q", actions)
+	}
+}
+
+func TestCallToolRetryCancellationUsesAPIClassification(t *testing.T) {
+	client := NewClient(&http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+		return &http.Response{StatusCode: http.StatusServiceUnavailable, Body: http.NoBody}, nil
+	})})
+	client.MaxRetries = 1
+	client.sleep = func(context.Context, time.Duration) error { return context.Canceled }
+
+	_, err := client.CallTool(context.Background(), "https://mcp.dingtalk.com/server", "search_messages", nil)
+	if err == nil {
+		t.Fatal("CallTool() error = nil, want retry cancellation")
+	}
+
+	var typed *apperrors.Error
+	if !errors.As(err, &typed) {
+		t.Fatalf("CallTool() error = %T, want *errors.Error", err)
+	}
+	if typed.Category != apperrors.CategoryAPI {
+		t.Fatalf("Category = %q, want %q", typed.Category, apperrors.CategoryAPI)
+	}
+	if typed.Operation != "tools/call" {
+		t.Fatalf("Operation = %q, want tools/call", typed.Operation)
+	}
+	if typed.Reason != "request_cancelled" {
+		t.Fatalf("Reason = %q, want request_cancelled", typed.Reason)
+	}
+	actions := strings.Join(typed.Actions, "\n")
+	if strings.Contains(actions, "internal/syncdata") || strings.Contains(actions, "sync-oss") {
+		t.Fatalf("runtime retry cancellation contains discovery-only actions: %q", actions)
+	}
+}
+
+func TestDiscoveryRequestFailuresKeepDiscoveryClassification(t *testing.T) {
+	tests := []struct {
+		name       string
+		failure    error
+		operation  string
+		wantReason string
+		call       func(*Client) error
+	}{
+		{
+			name:       "initialize connection refused",
+			failure:    errors.New("dial tcp: connection refused"),
+			operation:  "initialize",
+			wantReason: "connection_refused",
+			call: func(client *Client) error {
+				_, err := client.Initialize(context.Background(), "https://mcp.dingtalk.com/server")
+				return err
+			},
+		},
+		{
+			name:       "tools list dns failure",
+			failure:    errors.New("dial tcp: lookup mcp.dingtalk.com: no such host"),
+			operation:  "tools/list",
+			wantReason: "dns_resolution_failed",
+			call: func(client *Client) error {
+				_, err := client.ListTools(context.Background(), "https://mcp.dingtalk.com/server")
+				return err
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			client := NewClient(&http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+				return nil, tt.failure
+			})})
+			client.MaxRetries = 0
+
+			err := tt.call(client)
+			if err == nil {
+				t.Fatal("request error = nil, want failure")
+			}
+			var typed *apperrors.Error
+			if !errors.As(err, &typed) {
+				t.Fatalf("request error = %T, want *errors.Error", err)
+			}
+			if typed.Category != apperrors.CategoryDiscovery {
+				t.Fatalf("Category = %q, want %q", typed.Category, apperrors.CategoryDiscovery)
+			}
+			if typed.Operation != tt.operation {
+				t.Fatalf("Operation = %q, want %q", typed.Operation, tt.operation)
+			}
+			if typed.Reason != tt.wantReason {
+				t.Fatalf("Reason = %q, want %q", typed.Reason, tt.wantReason)
+			}
+		})
+	}
+}
+
 type roundTripFunc func(*http.Request) (*http.Response, error)
 
 func (f roundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) {
diff --git a/scripts/policy/open-source-audit.sh b/scripts/policy/open-source-audit.sh
index 819ebf44..07b93d7a 100755
--- a/scripts/policy/open-source-audit.sh
+++ b/scripts/policy/open-source-audit.sh
@@ -36,6 +36,14 @@ require_no_match() {
   fi
 }
 
+require_no_tracked_path() {
+  path="$1"
+  if git ls-files -- "$path" | grep -q .; then
+    git ls-files -- "$path" >&2
+    err "found forbidden tracked path: $path"
+  fi
+}
+
 require_heading_if_linked() {
   label="$1"
   heading="$2"
@@ -62,6 +70,10 @@ require_no_match 'scripts/test\.sh|scripts/check-semantic-fixtures\.sh|scripts/r
 require_no_match 'scripts/check-semantic-fixtures\.py|test/semantic/cli_to_mcp|test/semantic/model_to_cli|python3 -m pytest test/semantic' CONTRIBUTING.md
 require_no_match '\.github/CODEOWNERS|Tag-triggered GitHub Release asset publishing' CHANGELOG.md
 
+require_no_tracked_path 'auto-test'
+require_no_tracked_path 'eval-runs'
+require_no_tracked_path '.qoder'
+
 require_heading_if_linked '[Environment Variables](#environment-variables)' '## Environment Variables'
 require_heading_if_linked '[Exit Codes](#exit-codes)' '## Exit Codes'