Lark / Feishu Skills
The 5-skill bundle for Lark — Skill Maker, Events, Wiki, the 19-skill CLI suite, and the standup workflow that runs daily.
What's in this pack
This pack groups the five Lark / Feishu skills that together cover the full surface a Claude Code agent needs to read and write inside the Lark Suite — chat, wiki, base, approvals, and the daily standup ritual most teams run on Lark.
| # | Skill | Surface | What it does |
|---|---|---|---|
| 1 | feishu-skill-maker | meta | Generates new Feishu API skills from an OpenAPI snippet |
| 2 | feishu-events | webhook | Subscribes to message / approval / calendar events |
| 3 | feishu-wiki-search | Wiki + Docs | Full-text search across spaces with permission scoping |
| 4 | feishu-cli (19 commands) | shell | One-line wrappers for chat, file, base, calendar, group |
| 5 | daily-standup-workflow | workflow | Pulls yesterday's commits + today's calendar, posts a card |
The bundle assumes you have a tenant-level Lark / Feishu app and an app_id / app_secret pair. Skill Maker and the CLI suite work for both Lark (international) and Feishu (China) — the API surfaces are mirror images, only the host changes.
Why Lark skills are unusual
Most "AI + Lark" integrations stop at "post a webhook." This pack inverts the relationship: Claude Code becomes a Lark client, with bidirectional reads. The agent can browse a Wiki space, pull a Bitable record, file an approval, and reply in a Messenger thread — all from inside its normal session.
The reason this matters: Lark / Feishu hold the canonical document store for most China-based and APAC engineering teams. When the agent can read those docs natively, you stop pasting them in.
Install in one command
# Install the entire pack into the current Claude Code project
tokrepo install pack/lark-feishu-skills
# Or pick a single skill
tokrepo install feishu-cli
tokrepo install daily-standup-workflow
Once installed, each skill is a SKILL.md under .claude/skills/ plus the helper scripts the skill calls (Python or Node depending on the upstream). On first use, the skill prompts for LARK_APP_ID / LARK_APP_SECRET and writes them to a local .env it adds to .gitignore.
Common pitfalls
- Lark vs Feishu hosts. Same APIs, different domains (
open.larksuite.comvsopen.feishu.cn). SetLARK_HOSTper project, not globally. - Permission scopes. Wiki search and approval skills need scopes the default app doesn't have. The skill's pre-flight check lists exactly which scopes to add and links the admin console.
- Webhook signature verification.
feishu-eventsenforces signature verification by default. Disabling it for local dev means production traffic can spoof the agent — keep it on, use a tunnel like Cloudflared. - CLI rate limits. The 19-command CLI suite respects Lark's per-tenant QPS but does not coordinate across multiple Claude sessions. If you run two agents at once, lower the concurrency in
feishu-cli config. - Standup card formatting. The daily-standup workflow ships a default card layout; override
STANDUP_CARD_TEMPLATEif your team uses a custom template.
Relationship to other packs
These skills are designed to compose with two other packs:
- Anthropic Builders ships the Skill format itself; this pack uses it.
- Workflow Orchestration (n8n, Prefect) gives you an alternative when the standup workflow needs to run on a schedule independent of Claude Code being open.
If your stack is Slack + Notion instead, the surface mapping is mostly mechanical — but expect a re-write of the Wiki search skill, since Notion's permission model is fundamentally different from Lark's space ACLs.
What this pack alone isn't enough for
This pack covers tenant APIs but leaves out three adjacent concerns. First, OAuth user-flow login: the bundle uses tenant-level credentials, so per-user authorization (where the agent acts as a specific employee) needs an additional skill that wraps the Authorization Code flow. Second, voice and meeting transcripts: Lark Meetings has its own subscription model and is not in scope here. Third, mobile-app push notifications: those go through Lark Push, which requires extra setup beyond webhook delivery.
For each gap, the skill registry pages link to the appropriate community add-on. The pack stays narrow on purpose — five well-tested skills beat fifteen half-tested ones, and the standup workflow has been the highest-value asset for teams who install everything.
5 assets in this pack
Frequently asked questions
Is the pack free?
Yes — every skill is open source and the TokRepo install is free. You only pay for the Lark / Feishu plan you already have. The skills do not introduce a TokRepo-side proxy or token; they call open.larksuite.com or open.feishu.cn directly with your tenant credentials. Audit-friendly for security review.
How does this compare to a generic Lark webhook bot?
A webhook bot only pushes messages out. This pack adds bidirectional reads: the agent can pull a Wiki page, query a Bitable, fetch chat history, and respond contextually. The Skill Maker also lets you scaffold new endpoints from an OpenAPI snippet in under a minute, so you grow coverage without writing client code.
Works with Claude Code or Cursor?
Native fit is Claude Code — the skills follow Anthropic's SKILL.md format and ship as a folder under .claude/skills/. Cursor users get the CLI suite (which is just shell wrappers) and the daily-standup workflow as a script. Wiki search and Skill Maker need an Anthropic-style skill loader, so on Cursor you'd run them via the Claude Code CLI in a sub-shell.
Diff vs the Slack / Notion equivalents?
Slack and Notion split the surface across two products; Lark fuses chat + docs + base + calendar in one tenant. That means one OAuth, one rate-limit budget, one signature scheme — the skills exploit this by sharing a single auth helper. If your team uses Slack and Notion separately you'll want their respective MCP servers, not this pack.
Operational gotcha with the standup workflow?
The workflow assumes a chat group ID and pulls commits from the GitHub repos listed in standup-config.yaml. If that file ages out (someone leaves, a repo is renamed) the card silently shows partial data. Set up a weekly check that the configured repos still resolve — the workflow ships a --validate flag for exactly this.
12 packs · 80+ hand-picked assets
Browse every curated bundle on the home page
Back to all packs