Installable skillMIT

Feishu / Lark

Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.

BY HmbownUPDATED Oct 30, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: feishu-e3e20a53c526
description: Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.
---

# Feishu / Lark

Use this skill when the user asks for Feishu, Lark, or "飞书" integration work.

## Ground Rules

- Feishu China APIs use `open.feishu.cn`; Lark international APIs use
  `open.larksuite.com`.
- Never hardcode app secrets, webhook secrets, tenant tokens, or user tokens.
  Use environment variables such as `FEISHU_APP_ID`,
  `FEISHU_APP_SECRET`, `FEISHU_WEBHOOK_URL`, and
  `FEISHU_WEBHOOK_SECRET`.
- If credentials are unavailable, produce setup instructions or a local stub
  instead of pretending the integration is live.

## Common Use Cases

- Bot webhook messages
- App access token and tenant access token flows
- Docs, Sheets, Wiki, and Bitable reads/writes
- Approval or workflow status updates
- Feishu/Lark MCP server configuration

## Workflow

1. Clarify whether the target is Feishu or Lark.
2. Identify the credential type: webhook, internal app, marketplace app, or
   OAuth user token.
3. Prefer official OpenAPI endpoints and signed webhooks when secrets are
   configured.
4. For MCP, build or configure a server that exposes narrow tools such as
   `send_message`, `read_doc`, `append_sheet_row`, or `query_bitable`.
5. Register the MCP server with `codewhale mcp add`, then run
   `codewhale mcp validate` and `codewhale mcp tools`.
6. Verify with a dry run, sandbox document, or read-back call before sending
   externally visible messages.

Ask for confirmation before sending messages, writing production documents, or
changing approval/workflow state.
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.