当需要用 lark-cli 操作飞书多维表格（Base）时调用：适用于建表、字段管理、记录读写、视图配置、历史查询，以及角色/表单/仪表盘管理/工作流；也适用于把旧的 +table / +field / +record 写法改成当前命令写法。涉及字段设计、公式字段、查找引用、跨表计算、行级派生指标、数据分析需求时也必须使用本 skill。

2026-05-0212

lark-calendar

baibairui/WoClaw

飞书日历（calendar）：提供日历与日程（会议）的全面管理能力。核心场景包括：查看/搜索日程、创建/更新日程、管理参会人、查询忙闲状态及推荐空闲时段、查询/搜索与预定会议室。注意：涉及【预约日程/会议】或【查询/预定会议室】时，必须先读取 references/lark-calendar-schedule-meeting.md 工作流！高频操作请优先使用 Shortcuts：+agenda（快速概览今日/近期行程）、+create（创建日程并按需邀请参会人及预定会议室）、+freebusy（查询用户主日历的忙闲信息和rsvp的状态）、+rsvp（回复日程邀请）

2026-05-0212

lark-doc

baibairui/WoClaw

飞书云文档：创建和编辑飞书文档。从 Markdown 创建文档、获取文档内容、更新文档（追加/覆盖/替换/插入/删除）、上传和下载文档中的图片和文件、搜索云空间文档。当用户需要创建或编辑飞书文档、读取文档内容、在文档中插入图片、搜索云空间文档时使用；如果用户是想按名称或关键词先定位电子表格、报表等云空间对象，也优先使用本 skill 的 docs +search 做资源发现。

2026-05-0212

name	gateway-browser
description	Use when tasks require operating web pages. Run the bundled browser script in this skill so browser actions stay observable and reversible.

Gateway Browser Skill

When the user asks for browser operations, run the bundled browser script in this skill.

Workflow:

Read current page state with node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs snapshot before deciding the next action.
Execute one minimal action at a time (click/type/select-option/navigate/wait-for).
Re-run snapshot after each key action; refs are not stable across navigations.
If needed, use screenshot, tabs, evaluate, or recording commands to collect evidence.
Report action, evidence, result, and next step.

Environment:

GATEWAY_BROWSER_API_BASE
GATEWAY_INTERNAL_API_TOKEN

Examples:

node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs snapshot
node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs navigate --url "https://example.com"
node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs click --ref e1
node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs type --ref e2 --text "hello" --submit true
node ./.codex/skills/gateway-browser/scripts/gateway-browser.mjs tabs --action list

Report format:

Action: what changed on the page in this step.
Evidence: snapshot/screenshot/console/network findings that support the conclusion.
Result: success, blocked state, or failure reason.
Next step: the next minimal action or the exact user takeover request.

Status templates:

In progress: report the latest action, current evidence, and immediate next step.
Blocked: report the blocker, risk, and the exact decision needed from the user.
Handoff: report why user takeover is required, what state is preserved, and how to resume.
Done: report what was completed, the final result/artifact, and any follow-up suggestion.

Stop conditions:

Stop when the page intent is ambiguous, multiple similar targets exist, or the expected page state did not appear.
Stop when a modal, redirect, or permission prompt changes the task scope unexpectedly.
Stop when the action would send external data, upload files, or submit content the user did not explicitly approve.
When stopping, report the last confirmed page state and the exact decision needed from the user.

Rules:

Reuse existing tabs whenever possible.
Prefer visible, reversible actions over hidden shortcuts.
Do not run Playwright directly or invent another browser wrapper; use only the script bundled in this skill.
On login/OTP/captcha/payment confirmation, request user takeover.
Before submit/delete/publish/payment or other irreversible actions, capture evidence and confirm intent if the user did not state it explicitly.
If the page is visually unclear or the user asks what is on screen, capture a screenshot instead of guessing from stale refs.
If an action fails twice, stop and ask for user decision.
If a click/type fails, refresh the snapshot first, then inspect console/network before retrying.
During user takeover, keep the current tab/state intact and report the exact resume point.
Treat logged-in browser profiles, cookies, and storage as sensitive data.
Avoid arbitrary page-context JS unless it is necessary to inspect or unblock the task.