菜单
Installs a ready-made agent pack into an existing team repo, checks it does not break the agents already there, and reports what it needs to work. Trigger on /add-pack, install a pack, or add the <name> pack.
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/AutomatedMarketer/agent-team-os --skill add-pack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。
Checks an agent team repo and reports what works, what is stale, and what was never set up. Trigger on /audit, audit my team, check my setup, or what is broken.
Compares the jobs a repo declares against the routines that actually exist, and switches on only the ones the owner approved - each with a reason. Trigger on /arm, switch it on, make it run, arm my jobs, why is nothing running, or which of my jobs are real.
Reads ledger.yml and the team catalogue, and writes proposals.yml - what the owner's own numbers say their team should be. Every proposal cites their words, their number and a thing that already exists. Trigger on /match, what should my team be, who should I hire, what can you take off my week, or after /ledger.
基于 SOC 职业分类
正在显示 SKILL.md
| name | add-pack |
| description | Installs a ready-made agent pack into an existing team repo, checks it does not break the agents already there, and reports what it needs to work. Trigger on /add-pack, install a pack, or add the <name> pack. |
A pack is one or more agents plus their skills, already built and audited. Installing one should take a minute and change nothing that already works.
Show the user three things about the pack and wait for a go:
| Show | Because |
|---|---|
| What it does, in one line | They should know before it lands |
| Which model each agent uses, and how often it is meant to run | It comes out of their daily allowance |
| What it needs connected | A pack that needs a CRM is useless to someone without one |
If the pack needs something they do not have, say so plainly and ask whether they still want it installed. A pack that says "I cannot work until you connect X" on every run is worse than no pack.
git status must be clean. If it is not, commit or stash before
going further — a half-installed pack on top of uncommitted work is hard to unpick..claude/agents/..claude/skills/.agents/<slug>/knowledge/, fill markers intact.AGENT_SLUGS in scripts/lib/run-log.mjs, AGENT_SPECS in
scripts/lib/agents.mjs, IMPLEMENTED in tests/agents.test.mjs.CLAUDE.md and .claude/rules/routing.md.node scripts/build-model-card.mjs.node scripts/prompt-audit.mjs
npm test
Both clean. If the audit reports findings inside the pack's own files, stop and report it — that is a defect in the pack, not something to work around locally.
Then confirm nothing already there broke:
"Ask your research agent for anything. Then ask your content agent for anything. Both should behave exactly as they did before."
## <Pack name> installed
**Added:** <agents>, <skills>
**Models:** <slug> on <alias>, ...
**Needs connected:** <list, or "nothing">
**Suggested schedule:** <what, how often, and what that costs against the daily cap>
Your existing agents were run after install and behave unchanged.
Removal is the reverse, and it has to leave the repo exactly as it was:
.claude/agents/ and .claude/skills/npm testRun logs the pack's agents wrote stay. They are history, and history is not undone by an uninstall.