setup
Verify repo-hygiene's external prerequisites on this machine and report the effective destructive-guard toggle and the scope it actually applies at. The prerequisites are `git`, which the scan, git, stash, and tree tiers and the tracked-file guarantee all rest on, `node`, which launches the destructive guard, and the optional `ghq` the fleet batch actions enumerate repositories from. Use when: 'set up repo-hygiene', 'configure repo-hygiene', 'is repo-hygiene working', 'is the destructive guard on', 'why did tree-batch find no repos', or before a first clean on a new machine. Check-only: verifies, reports, and points at each remediation; installs nothing and there is nothing setup may write here. Re-runnable and safe.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 30
- Forks
- 7
01 前言 Front Matter
- shell
- bash
- description
- Verify repo-hygiene's external prerequisites on this machine and report the effective destructive-guard toggle and the scope it actually applies at. The prerequisites are `git`, which the scan, git, stash, and tree tiers and the tracked-file guarantee all rest on, `node`, which launches the destructive guard, and the optional `ghq` the fleet batch actions enumerate repositories from. Use when: 'set up repo-hygiene', 'configure repo-hygiene', 'is repo-hygiene working', 'is the destructive guard on', 'why did tree-batch find no repos', or before a first clean on a new machine. Check-only: verifies, reports, and points at each remediation; installs nothing and there is nothing setup may write here. Re-runnable and safe.
- allowed-tools
- [ "Bash(command -v ghq*)" ]
- argument-hint
- [check]
- user-invocable
- true
- disable-model-invocation
- true
03 SKILL.md 正文
Pre-computed context
check's ghq probe ran at load time. Read this row instead of re-issuing it; it shows
the tool's path when present, or absent when missing:
ghq: !command -v ghq 2>/dev/null || echo absent
A row reading [shell command execution disabled by policy] carries no result: run that tool's
command -v probe via Bash instead.
- Pointer: for the placeholder a skipped load-time command leaves, fetch skills § Inject dynamic context live.
- As of: 2026-11-11
- Recheck trigger: the placeholder text changes.
git is probed in the body, not here (gather-block convention,
"The pre-compute block runs as one shell invocation").
Purpose
Check-only setup under the Check-only carve-out (docs/plugin-philosophy.md "Setup is explicit
and repeatable" in the marketplace repository): this plugin's configuration contains no
writable artifact, so check inspects, reports, and points at each remediation, and no apply is
offered because there is nothing it could conformingly write. The warrant is the carve-out's
external-prerequisites class: git, which every git-touching tier of /repo-hygiene:clean and
the tracked-file safety guarantee depend on, node, which launches the destructive guard, and the
optional ghq the fleet batch actions
enumerate repositories from. None is visible to a native configuration prompt; each is
verifiable only. The clean_destructive_guard_enabled option is a native userConfig toggle whose only
stored home is the pluginConfigs this contract forbids setup to write.
Action routing: no argument or check runs the check. Non-interactive, never prompts.
check (read-only)
The clean skill and its bundled scripts are the single source of truth for what each tier requires.
Read it first, with the Read tool, by the paths the steps below name: ${CLAUDE_PLUGIN_ROOT}/skills/clean/SKILL.md first, then each script a step names. Probe what it actually does, don't recite this file.
Never search the plugin directory: no Glob, Grep, find, ls or cd in ${CLAUDE_PLUGIN_ROOT}, and no shell head or cat of its files. The files below are all it needs. Then read the
pre-computed ghq row, run each remaining probe as its own Bash call with nothing chained to it, and report a PASS/FAIL/INFO
table with one remediation line per FAIL. Do not modify anything.
Install nothing, and run no mutating tier.
-
gitonPATH.command -v git, and report the resolved path and version. FAIL when absent, and state what is lost rather than a blanket "the plugin is broken":scanresolves the repository through git and reportsnot a git repositorywhen it cannot; thegit,stash,tree, andtree-batchtiers are git operations outright.- The any git-tracked file is off-limits guarantee is enforced against the index by the
shared
${CLAUDE_PLUGIN_ROOT}/skills/clean/scripts/lib/clean-common.sh(Read that file to confirm it) (git ls-files --error-unmatch), so even thecachesandbuildtiers reach git through it, as does restoring a tracked file deleted by reparse-point traversal during atreeclean. No tier is safe to run without git.
-
ghq(optional). The pre-computedghqrow. Present: INFO,ghq list -pcan feedtree-batchand the other*-batchactions. Absent: INFO, not a defect, the batch actions still take--repo(repeatable, glob-expanded) and--repos-from FILE|-; only theghq-derived enumeration is unavailable. Report this rather than letting an empty repo list look like a bug. -
A POSIX shell for the bundled scripts. Every tier script and the destructive guard are
bash. On Windows that means Git Bash must be present. The guard is exec form:commandisnode, and${CLAUDE_PLUGIN_ROOT}/hooks/exec-bash.mjs(Read it to confirm) finds bash (on WindowsCLAUDE_CODE_GIT_BASH_PATH, thenGit\bin\bash.exe, thenPATH, neverSystem32\bash.exe; <!-- portability-ok: Windows path, not a shell regex --> elsewherePATH, then/bin/bashand/usr/bin/bash).- Pointer: for how an exec-form hook launches its command, fetch https://code.claude.com/docs/en/hooks#exec-form-and-shell-form live.
Report the shell as INFO on Unix; FAIL on Windows when no Git Bash resolves, since the scripts and the guard alike cannot launch.
-
nodeonPATH.command -v node, and report the resolved path and version. Probe it through Bash, so it works without the launcher. FAIL when absent: every hook row runsnode hooks/exec-bash.mjs, so without it the guard does not launch and is not enforced.- Pointer: for whether Claude Code's install uses Node at runtime, fetch https://code.claude.com/docs/en/setup#install-with-npm live.
-
Destructive-guard registration and toggle. INFO, and be precise about where the guard lives, because the answer is the reason it is session-scoped:
-
It registers from the frontmatter
hooks:block in${CLAUDE_PLUGIN_ROOT}/skills/clean/SKILL.md, not from a plugin-levelhooks/hooks.json, which this plugin does not ship. Claude Code arms it once/repo-hygiene:cleanis invoked and keeps it armed for the rest of that session, so a session that never invoked the skill has no guard and nothing is wrong with that. Read that SKILL.md for the block (it runs${CLAUDE_PLUGIN_ROOT}/skills/clean/scripts/destructive-guard.sh); never look for ahooks.jsonor listhooks/. -
Report the effective
${user_config.clean_destructive_guard_enabled}(an unexpanded token or an empty value means the manifest defaulttrue). Read it from this text as rendered; never echo it through Bash, where an unexpanded token is invalid shell syntax and the call is refused. A change made now is observed only in a fresh session.- Pointer: for when a saved value reaches skill content, fetch plugins manifest reference § Reference a saved value live.
-
The option is user-scoped: plugin option values are read from user,
--settings, and managed settings only, never from a project's.claude/settings.json. So there is no per-repository value of this toggle. To vary the behavior for one repository, enable or disable the plugin in that project'senabledPluginsinstead.
-
Remediation guidance (printed by check; the operator applies it)
The prerequisites are system tools and the one option lives in Claude Code's native
plugin configuration (Check-only carve-out, external-prerequisites and native-userConfig
classes), so check closes by pointing at each resolution rather than writing. Re-running it
after everything passes changes nothing and reports "already configured":
-
Missing
git: the platform's own install channel (https://git-scm.com/downloads). This plugin never downloads a tool. -
Missing
ghqand fleet actions wanted: install it (https://github.com/x-motemen/ghq), or keep using--repo/--repos-fromand say so. This is a convenience, not a blocker. -
Missing
node: install Node.js (https://nodejs.org/en/download) and put it onPATH. Until then the destructive guard is not enforced. -
Missing Git Bash on Windows: install Git for Windows; nothing in this plugin runs without it.
-
Toggle off (or on): set
clean_destructive_guard_enabledtotrueorfalsethrough Claude Code's native flow, interactively with/plugin configure repo-hygiene@<marketplace>or by a headless route, as the marketplace's plugin-reconfiguration convention gives them in § The routes (it owns the verified-version record). Apply the scope and failure-signal caveats in its § Caveats every setup skill's short form carries. Read both sections and print what they give, the fallback route, its behavior on an already-installed plugin and the scope to pass included, rather than only naming them, and leave out the verified-version record. Never uninstall to reconfigure; that section links what an uninstall deletes. This skill never writes user settings orpluginConfigs. Afterwards reruncheckin a fresh session, since a same-sessioncheckcan still show the old value, and report the observed effective toggle value, never an unobserved change.- Pointer: for the headless command's flags, how it reports a saved or refused value, and when a saved value applies, fetch plugin CLI reference § plugin configure live.
What this skill does NOT do
- Scan, clean, prune, or reset anything. Those are
/repo-hygiene:clean's tiers, with their own dry-run-first and confirmation contract. - Install
git,ghq, or any tool. Guidance only. - Write the plugin cache, Claude Code user settings, or
pluginConfigs. Nor any other Claude Code settings file.
04 同仓库更多 Skills
track
Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.
reviewer-media
Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from /review:explain-change's digest, has /explainer-video:produce render it and hands the manifest to /source-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.
route
Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.
route
Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.
principles
Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is /architecture:improve.
write
Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').