DingTalk Workspace CLI (dws)
dws — DingTalk Workspace on the command line, built for humans and AI agents.
中文版 · English · Reference · Changelog
> [!IMPORTANT]
> **Co-creation Phase**: This project accesses DingTalk enterprise data and requires enterprise admin authorization. Please join the DingTalk DWS co-creation group to complete whitelist configuration. See [Getting Started](#getting-started) below.
>
> 
Table of Contents
- [Why dws?](#why-dws)
- [Installation](#installation)
- [Getting Started](#getting-started)
- [Quick Start](#quick-start)
- [Using with Agents](#using-with-agents)
- [New Features](#new-features)
- [Key Services](#key-services)
- [Security by Design](#security-by-design)
- [AI Agent Skills](#ai-agent-skills)
- [Reference & Docs](#reference--docs)
- [Contributing](#contributing)
---
Why dws?
- **For humans** — `--help` for usage, `--dry-run` to preview requests, `-f table/json/raw` for output formats.
- **For AI agents** — structured JSON responses + built-in Agent Skills, ready out of the box.
- **For enterprise admins** — zero-trust architecture: OAuth device-flow auth + domain allowlisting + least-privilege scoping. **Not a single byte can bypass authentication and audit.**
## Installation
**macOS / Linux:**
```bash
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.sh | sh
```
**Windows (PowerShell):**
```powershell
irm https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.ps1 | iex
```
Other install methods
**Pre-built binary**: download from [GitHub Releases](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/releases).
**Build from source**:
```bash
git clone https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli.git
cd dingtalk-workspace-cli
go build -o dws ./cmd # build to current directory
cp dws ~/.local/bin/ # install to PATH
```
> Requires Go 1.25+. Use `make package` to cross-compile for all platforms (macOS / Linux / Windows x amd64 / arm64).
## Getting Started
### Step 1: Create a DingTalk Application
Go to the [Open Platform Console](https://open-dev.dingtalk.com/fe/app?hash=%23%2Fcorp%2Fapp#/corp/app). Under "Internal Enterprise Apps - DingTalk Apps", click **Create App**.
View screenshot
### Step 2: Configure Redirect URL
Go to app settings → **Security Settings**. Add the following redirect URLs and save:
```
http://127.0.0.1
https://login.dingtalk.com
```
> `http://127.0.0.1` is for local browser login; `https://login.dingtalk.com` is for `--device` device-flow login (Docker containers, remote servers, and other headless environments). We recommend configuring both.
View screenshot
### Step 3: Publish the Application
Click "App Release - Version Management & Release" to publish and go live.
View screenshot
### Step 4: Request Whitelist Access
Join the DingTalk DWS co-creation group and provide your **Client ID** and **admin confirmation** to complete whitelist setup.
### Step 5: Authenticate
```bash
dws auth login --client-id --client-secret
```
Or via environment variables:
```bash
export DWS_CLIENT_ID=
export DWS_CLIENT_SECRET=
dws auth login
```
> CLI flags take precedence over environment variables. Credentials are used for DingTalk's OAuth device flow.
## Quick Start
```bash
dws contact user search --keyword "Alice" # search contacts
dws calendar event list # list calendar events
dws todo task create --title "Quarterly report" --executors "" # create a todo
dws todo task list --dry-run # preview without executing
```
## Using with Agents
dws is designed as an AI-native CLI. Here's how to integrate it with your agent:
### 1. Setup
```bash
# Install dws
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install.sh | sh
# Configure auth via environment variables (recommended for agents, no interactive login)
export DWS_CLIENT_ID=
export DWS_CLIENT_SECRET=
dws auth login
```
### 2. Agent Invocation Patterns
```bash
# Use --yes to skip confirmation prompts (required for agents)
dws todo task create --title "Review PR" --yes
# Use --dry-run to preview operations (safe execution)
dws contact user search --keyword "test" --dry-run
# Use --jq to extract precisely (save tokens)
dws contact user search --keyword "test" --jq '.response.content'
# Use --fields to return only needed fields
dws calendar event list --fields response
```
### 3. Schema Discovery (Agent Self-Exploration)
Agents don't need pre-built knowledge of every command. Use `dws schema` to dynamically discover capabilities:
```bash
# Step 1: Discover all available products
dws schema --jq '.products[] | {id, tool_count: (.tools | length)}'
# Step 2: Inspect target tool's parameter schema
dws schema aitable.query_records --jq '.tool.input_schema'
# Step 3: Construct the correct call
dws aitable record query --base-id "xxx" --table-id "yyy" --limit 10
```
### 4. Error Tolerance
The smart correction engine is highly tolerant of common agent-generated mistakes:
| Agent Output | dws Auto-Corrects To |
|-----------|--------------|
| `--userId` | `--user-id` |
| `--limit100` | `--limit 100` |
| `--tabel-id` | `--table-id` |
| `--USER-ID` | `--user-id` |
| `--user_name` | `--user-name` |
### 5. Agent Skill Integration
After installing Agent Skills, tools like Claude Code / Cursor can use DingTalk capabilities directly:
```bash
# Install skills into current project
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-skills.sh | sh
```
> See the [AI Agent Skills](#ai-agent-skills) section for details.
## New Features
### Smart Input Correction
dws includes a built-in pipeline engine that auto-corrects common AI model parameter mistakes:
```bash
# Naming convention auto-conversion (camelCase / snake_case / UPPER -> kebab-case)
dws aitable record query --baseId x --tableId y # auto-corrected to --base-id --table-id
# Sticky argument splitting
dws contact user search --keyword "test" --timeout30 # auto-split to --timeout 30
# Fuzzy flag name matching
dws aitable record query --base-id x --tabel-id y # --tabel-id -> --table-id
# Value normalization (boolean / number / date / enum)
# "yes" -> true, "1,000" -> 1000, "2024/03/29" -> "2024-03-29", "ACTIVE" -> "active"
```
### JSON Field Selection & jq Filtering
Fine-grained output control to reduce token consumption:
```bash
# Return only specific fields
dws aitable record query --base-id x --table-id y --fields invocation,response
# Built-in jq expressions
dws aitable record query --base-id x --table-id y --jq '.invocation.params'
dws schema --jq '.products[] | {id, tools: (.tools | length)}'
```
### Schema Introspection
Agents can query parameter schemas before making calls:
```bash
dws schema # list all products and tools
dws schema aitable.query_records # view parameter schema
dws schema aitable.query_records --jq '.tool.input_schema.required' # view required fields
dws schema --jq '.products[].id' # extract all product IDs
```
## Key Services
| Service | Command | Description |
|---------|---------|-------------|
| Contact | `contact` | Users / departments |
| Chat | `chat` | Group management / members / bot messaging / webhook |
| Calendar | `calendar` | Events / meeting rooms / free-busy |
| Todo | `todo` | Task management |
| Approval | `approval` | Processes / forms / instances |
| Attendance | `attendance` | Clock-in / shifts / statistics |
| Ding | `ding` | DING messages / send / recall |
| Report | `report` | Reports / templates / statistics |
| AITable | `aitable` | AI table operations |
| Workbench | `workbench` | App query |
| DevDoc | `devdoc` | Open platform docs search |
Run `dws --help` for the full list, or `dws --help` for subcommands.
Coming soon
`doc` (documents) · `mail` (email) · `minutes` (AI transcription) · `drive` (cloud drive) · `conference` (video) · `tb` (Teambition) · `aiapp` (AI apps) · `live` (streaming) · `skill` (marketplace)
Security by Design
`dws` treats security as a first-class architectural concern, not an afterthought. **Credentials never touch disk, tokens never leave trusted domains, permissions never exceed grants, operations never escape audit** — every API call must pass through DingTalk Open Platform's authentication and audit chain, no exceptions.
For Developers
| Mechanism | Details |
|-----------|----------|
| **Encrypted token storage** | **PBKDF2 + AES-256-GCM** encryption, keyed by device physical MAC address; cross-platform Keychain/DPAPI integration provides additional protection — tokens cannot be decrypted on another machine |
| **Input security** | Path traversal protection (symlink resolution + working directory containment), CRLF injection blocking, Unicode visual spoofing filtering — prevents AI Agents from being tricked by malicious instructions |
| **Domain allowlist** | `DWS_TRUSTED_DOMAINS` defaults to `*.dingtalk.com`; bearer tokens are never sent to non-allowlisted domains |
| **HTTPS enforced** | All requests require TLS; HTTP only permitted for loopback during development |
| **Dry-run preview** | `--dry-run` shows call parameters without executing, preventing accidental mutations |
| **Zero credential persistence** | Client ID / Secret used in memory only — never written to config files or logs |
For Enterprise Admins
| Mechanism | Details |
|-----------|---------|
| **OAuth device-flow auth** | Users must authenticate through an admin-authorized DingTalk application |
| **Least-privilege scoping** | CLI can only invoke APIs granted to the application — no privilege escalation |
| **Allowlist gating** | Admin confirmation required during co-creation phase; self-service approval planned |
| **Full-chain audit** | Every data read/write passes through the DingTalk Open Platform API — enterprise admins can trace complete call logs in real time; no anomalous operation can hide |
For ISVs
| Mechanism | Details |
|-----------|---------|
| **Tenant data isolation** | Operates under authorized app identity; cross-tenant access is impossible |
| **Skill sandbox** | Agent Skills are Markdown documents (`SKILL.md`) — prompt descriptions only, no arbitrary code execution |
| **Zero blind spots** | Every API call during ISV–dws skill orchestration is forced through DingTalk Open Platform authentication — full call chain is traceable with no bypass path |
> Found a vulnerability? Report via [GitHub Security Advisories](https://github.com/DingTalk-Real-AI/dingtalk-workspace-cli/security/advisories/new). See [SECURITY.md](./SECURITY.md).
## AI Agent Skills
The repo ships Agent Skills (`SKILL.md` files) for every DingTalk product. The install script deploys them to `~/.agents/skills/dws` automatically.
```bash
# Install skills into the current project only
curl -fsSL https://raw.githubusercontent.com/DingTalk-Real-AI/dingtalk-workspace-cli/main/scripts/install-skills.sh | sh
```
> `install.sh` installs to `$HOME/.agents/skills/dws` (global); `install-skills.sh` installs to `./.agents/skills/dws` (current project).
### ISV Skill Integration
Author your own Agent Skills and orchestrate them with dws skills for cross-product workflows: **ISV Skill → dws Skill → DingTalk Open Platform API (enforced auth + full audit)**.
**Example**: A CRM Skill invokes the Calendar Skill to create a client meeting, then triggers the Todo Skill to assign follow-ups — the AI agent completes cross-system collaboration in a single conversation.
## Reference & Docs
- [Reference](./docs/reference.md) — environment variables, exit codes, output formats, shell completion
- [Architecture](./docs/architecture.md) — discovery-driven pipeline, IR, transport layer
- [Changelog](./CHANGELOG.md) — release history and migration notes
## Contributing
See [CONTRIBUTING.md](./CONTRIBUTING.md) for build instructions, testing, and development workflow.
## License
Apache-2.0