菜单
Write or improve a SKILL.md so it actually triggers and actually changes agent behavior. Use when the user says "create a skill", "turn this into a skill", "why isn't this skill triggering", or when compound-learnings decides a lesson deserves a skill.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/aiosbrain/aios-workspace --skill skill-author命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Manage the AIOS Linear board (the ONLY PM tool — Plane is retired). Use whenever updating, reading, or commenting on AIOS Linear issues (identifiers like AIO-72, AIO-75), the brain→PM projection-tracking epic, or the backlog. Triggers on "update AIO-NN", "the Linear board", "the AIOS board", "projection tracking issue", "/aios-linear". Provides a terse `linear.mjs` CLI so issue edits are one fast command instead of ad-hoc GraphQL one-liners. DO NOT use the Plane MCP for AIOS work — it is retired and its tools return huge irrelevant payloads.
Use YOUR own Slack to act as you — send messages and DMs to teammates, read your channels/DMs, and react — via the `slack` CLI (a zero-dependency tool that talks straight to the Slack Web API with your USER token). Messages post AS you and replies land in your own DMs. This is the INDIVIDUAL "act as me" connector — distinct from the team's read-only Slack history ingestion. Use when the user asks to message/DM a teammate on Slack, read a Slack thread/DM, or react. Requires Slack connected (run `aios connect slack`).
Verify that a fix or change actually did what it was for, before reporting that it works. Use before any "it works", "verified", "fixed", "confirmed", or "done" about something that was previously broken — and whenever a hook or a reviewer asks you to re-check a claim. Catches the failure where you measure a proxy (a neighbouring count, an absence of errors, a ratio) instead of the property the change exists to produce.
基于 SOC 职业分类
正在显示 SKILL.md
| name | skill-author |
| description | Write or improve a SKILL.md so it actually triggers and actually changes agent behavior. Use when the user says "create a skill", "turn this into a skill", "why isn't this skill triggering", or when compound-learnings decides a lesson deserves a skill. |
| source | Jesse Vincent (obra) — Superpowers skill-writing skill + pressure-testing; Anthropic skill-authoring guidance |
| am_pattern | C1 |
| triggers | ["write a skill","create a skill","new skill","skill.md","improve this skill","skill isn't triggering"] |
You are authoring a skill — a packaged procedure an agent loads when a task matches. A skill earns its place only if it (a) triggers at the right moment and (b) changes what the agent would otherwise do. Most bad skills fail one of these.
skills/<kebab-name>/SKILL.md # required
skills/<kebab-name>/<helper> # optional scripts/templates the skill references
Frontmatter: name (kebab, matches dir), description (the trigger — see below), and
in this pack also source: (provenance) and am_pattern: (maturity pattern tags).
The runtime matches tasks against the description, not the body. Include:
Weak: Helps with database work.
Strong: Guides schema migrations — use whenever adding/renaming a column or table. Never edits migration files directly; generates them via the migration tool.
Run the draft against 2–3 realistic scenarios with a fresh agent (subagent or new session), including one decision-conflict case where the easy path violates the skill ("the test is failing and the deadline is now — does the agent still refuse to skip it?"). If the agent ignores or misreads the skill, the fix is usually the description (didn't trigger) or the hard-rules section (didn't bite) — not more prose.
A skill that misfires twice gets rewritten or deleted. Skills are code: version them, review changes to them, and prune ruthlessly — a stale skill is worse than none.