DingTalk Workspace CLI (dws) exposes DingTalk's product suite — contacts, chat, calendar, todos, approvals, attendance, reports, AI tables, docs, drive, meeting minutes, mail, spreadsheets, and wiki — as a single command-line tool for both people and AI agents. Every command returns structured JSON, output can be narrowed with built-in jq expressions or field selection, and --dry-run previews a request without executing it.
Authentication uses an OAuth device flow against an admin-authorised DingTalk application. Tokens are stored encrypted and refreshed automatically. A custom-app mode with a client ID and secret covers CI/CD and ISV integration, profiles let one installation stay logged in to several organisations or accounts, and an export/import flow moves credentials between Linux sandboxes.
For agents, the tool ships Agent Skills in two layouts — one skill per product, or a single mono skill covering every product — placed in the ~/.agents/skills root read by Claude Code, Cursor, Codex, and other detected agent homes. A schema command exposes each leaf command's parameters, constraints, risk, and confirmation contract, and an input-correction pipeline normalises flag names, splits sticky arguments, fuzzy-matches typos, and normalises values that models commonly get wrong.
Beyond the product commands, dws event consume subscribes to personal DingTalk events (messages, reads, recalls, reactions, group lifecycle, approval and todo events) over a managed WebSocket and emits each one as an NDJSON line for event-driven agents. dws api calls any App-Token-capable DingTalk OpenAPI directly with managed tokens and optional auto-pagination. dws dev connect bridges a DingTalk robot to a local coding agent so that @-mentioning the bot in a chat is answered by the agent with per-conversation memory.
Security controls include a domain allowlist for bearer tokens (*.dingtalk.com by default), HTTPS enforcement, path-traversal and CRLF-injection blocking, Unicode spoofing filters, least-privilege scoping to the APIs granted to the application, and routing of every call through DingTalk Open Platform authentication and audit.
Features
- Product coverage: contact, chat, calendar, todo, approval, attendance, DING, report, AI tables, doc, drive, minutes, mail, sheet, wiki, dev-doc search, AI people search, and live streams
- Structured output: JSON, table, or raw formats with --jq filtering and --fields selection to limit tokens
- Agent Skills: per-product or mono skill layouts installed into a shared ~/.agents/skills root, with intent routing, error-code references, and 13 ready-made Python workflow scripts
- Schema introspection: dws schema returns a compact command contract for agent selection or a full export for CI and audits
- Input correction: converts camelCase, snake_case, and uppercase flags to kebab-case, splits --limit100 into --limit 100, and fixes misspelled flag names
- Event subscription: NDJSON stream of one-to-one and group messages, OA approval events, and todo lifecycle events, with shared connections and local status/stop commands
- Raw API access: direct calls to api.dingtalk.com and oapi.dingtalk.com with managed app tokens, JSON bodies from files or stdin, and multipart uploads
- Robot bridge: connects a DingTalk bot to a local AI CLI with /new and /clear session commands
- Multi-profile auth: several organisations and accounts per installation, selectable by corpId:userId or friendly names
- Self-upgrade: two-phase update with SHA256 verification, automatic backup, beta track, and rollback
- Security by design: encrypted token storage, domain allowlisting, dry-run previews, and full-chain audit through the Open Platform
