lark-attendance: Read Your Feishu Check-Ins From the CLI (729K)
Feishu’s lark-attendance skill is the thinnest file in the lark pack: one endpoint, permanently fixed params, self-only by construction. 728.7K installs.
The smallest file in the biggest skill pack does one thing and refuses to do anything else. Feishu’s lark-attendance skill (v1.0.0, inside larksuite/cli) is a 1,620-byte SKILL.md with exactly one API method — user_tasks.query, which returns your own check-in records. I read the file end to end and scraped its skills.sh detail page on 2026-09-27: 728,700 installs on the open.feishu.cn counter, first seen April 14, 2026, with a weekly band running in lockstep with the rest of the family (lark-im 730.1K, lark-calendar 730.0K, lark-sheets 729.7K, lark-attendance 728.7K) — the pack-level counter at work again.
One Endpoint, Two Parameters the Skill Refuses to Ask About
user_tasks.query takes two parameters that shape every call, and the skill hard-codes both of them. employee_type is pinned to "employee_no" and user_ids to [] — an empty array. The instruction is explicit: auto-fill these in every API call, never ask the user for them. The design intent is visible in the choice of empty array: this skill can only read the authenticated user’s own attendance records. There is no path in the skill to query a colleague’s clock-in history, because the parameter that would enable it is permanently fixed to “nobody else.”
That makes attendance the family’s safest skill by construction. Compare it with lark-approval, which carries a 6,969-byte SKILL.md plus 16 reference files to describe who may act on whom; attendance carries one read-only endpoint, one scope (attendance:task:readonly), and zero write operations. The pack’s answer to a PII-heavy domain is not more rules — it’s removing the verbs entirely.
The Schema-First Rule Applies to a One-Method Skill Too
The skill still enforces the family’s schema discipline. Before calling anything, the agent must run lark-cli schema attendance.user_tasks.query to inspect the --data / --params structure rather than guessing field formats — the same rule lark-calendar applies to its fourteen reference files, applied here to a single method. The CLI help lives at lark-cli attendance --help, and the whole skill assumes the shared auth setup: reading ../lark-shared/SKILL.md first is marked CRITICAL, because attendance data rides the same token and permission machinery as every other lark skill.
Why the Thinnest Skill Is Also the Most Telling
Look at the family’s install spread and you see one counter, not seven: lark-im 730.1K, lark-calendar 730.0K, lark-minutes 728.9K, lark-attendance 728.7K — all first seen the same week, all drifting together. Pack counting means popularity metrics here measure the platform, not the individual feature. What actually differentiates attendance is its minimalism: one byte-heavy method, two fixed parameters, zero references, one query-only scope. That’s the whole product — npx skills add https://open.feishu.cn/ --skill lark-attendance and your agent can answer “how many days did I work from home last month?” without ever being able to ask about anyone else’s.
The next time an agent framework ships a 500-line skill for something as personal as attendance, remember this one: the right amount of surface area for reading your own check-ins is a single read-only method with the “who” parameter welded shut.