stack-audit
Audits this Claude Code stack for things worth acting on — plugins and skills never invoked, version pins that look current but are stale, plaintext credentials in config, and orphaned state. Use for "what can I prune", "is anything out of date", "audit my setup", "what is costing me context".
- License 许可
- MIT
- 所属仓库
- Deutrix/ccatlas
- 最近更新
- 2026-08-15
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- stack-audit
- description
- Audits this Claude Code stack for things worth acting on — plugins and skills never invoked, version pins that look current but are stale, plaintext credentials in config, and orphaned state. Use for "what can I prune", "is anything out of date", "audit my setup", "what is costing me context".
02 资源文件 1
- 参考文档reference.md1.8 KB
03 SKILL.md 正文
Stack audit
Three commands, read in this order.
ccatlas doctor --json # findings, each with a severity and a fix command
ccatlas updates --json # stale pins, upgrades, marketplace staleness
ccatlas usage --json # invocation counts and the never-invoked list
The finding no other tool shows
updates.data.stalePins — the version string has not moved but the source
has, so /plugin update reports already at the latest version while old code
runs. Lead with these when present.
Before recommending a prune
usage returns {available:false, reason} when the transcript layer could not
be read. If it does, recommend nothing. "I could not read your usage" and
"you have not used these" are the same shape and opposite advice, and the
second gets acted on by deleting.
Also check data.unused[].passiveCost. Absent means nobody measured it — not
that it is free. Rank by cost only where a cost exists.
Presenting fixes
Every doctor finding carries fixCommand, or omits it when no single command
would work. Quote it verbatim; never invent one. Read reference.md for
the finding codes, severity meanings and the skipped list — checks that did
not run, which must be reported rather than read as a clean bill of health.
04 同仓库更多 Skills
stack-overview
Answers questions about what is installed in this Claude Code stack — which plugins, skills, agents, commands and MCP servers exist, where each came from, and whether the CLI and config files agree. Use for "what do I have installed", "where did this skill come from", "which plugin owns X".
stack-migrate
Walks a Claude Code stack migration to another machine — exporting a portable bundle, and producing a DRY-RUN plan for importing one. Use for "move my setup", "export my stack", "what would this bundle install". Never applies a remote bundle; that always needs the user.
synthetic-bravo-skill
SYNTHETIC FIXTURE — second skill contributed by the synthetic @skills-dir plugin. Deliberately shares its name with a personal skill so shadowing detection (T1.7) has a case.
synthetic-alpha-skill
SYNTHETIC FIXTURE — first skill contributed by the synthetic @skills-dir plugin. Its owner is the plugin, not the personal skills directory it happens to live under.
synthetic-misplaced-manifest
SYNTHETIC FIXTURE — a personal skill with a plugin.json at the directory ROOT rather than inside .claude-plugin/. Must be classified personal, not plugin.
synthetic-decoy-only
SYNTHETIC FIXTURE — a personal skill whose directory carries a FOREIGN tool manifest and no Claude Code manifest. Must be classified personal, not plugin.