AGENT SKILL

setup

Verify bash-format's prerequisites and configuration here, and register its git pre-commit entry. Use when: 'set up bash-format', 'configure bash-format', 'is bash-format working', shell lint or formatting silently isn't happening, or a hook reported a missing prerequisite or drifted git hook. Actions: check (read-only verification, default) | apply | apply install-git-hooks [scope=local|shared]. Re-runnable.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 30
Forks
7

01 前言 Front Matter

shell
bash
description
Verify bash-format's prerequisites and configuration here, and register its git pre-commit entry. Use when: 'set up bash-format', 'configure bash-format', 'is bash-format working', shell lint or formatting silently isn't happening, or a hook reported a missing prerequisite or drifted git hook. Actions: check (read-only verification, default) | apply | apply install-git-hooks [scope=local|shared]. Re-runnable.
allowed-tools
[ "Bash(command -v jq*)", "Bash(command -v node*)", "Bash(command -v shellcheck*)", "Bash(command -v shfmt*)", "Bash(node \"${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs\" verify:*)", "Bash(node ${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs verify:*)", "Bash(node \"${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs\" plan:*)", "Bash(node ${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs plan:*)" ]
argument-hint
[check|apply [install-git-hooks [scope=local|shared]]]
user-invocable
true
disable-model-invocation
true

02 资源文件 1

03 SKILL.md 正文

Pre-computed context

check's tool probes ran at load time. Read these rows instead of re-issuing them; each shows the tool's path when present, or absent when missing:

  • jq: !command -v jq 2>/dev/null || echo absent
  • node: !command -v node 2>/dev/null || echo absent
  • shellcheck: !command -v shellcheck 2>/dev/null || echo absent
  • shfmt: !command -v shfmt 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.

Purpose

Thin check-centric setup per the uniform setup contract (docs/plugin-philosophy.md "Setup is explicit and repeatable" in the marketplace repository): check inspects and reports, apply resolves. The only consumer-project files this plugin owns are its git-hook entry copies, which apply install-git-hooks writes. Linting rules come from the repository's own .shellcheckrc, formatting from its .editorconfig, and the tunables are the native userConfig options (the on/off toggle, the gitignored-file switch and the end-of-turn check). Every prerequisite is a PATH binary the plugin never bundles, and the plugin never installs system packages, so the rest of apply is guidance-only. It never modifies user settings or the plugin cache.

Action routing: no argument or check runs the check; apply runs the check first, then offers remediation guidance. Both are non-interactive. Never prompt when the action is given. apply install-git-hooks asks only where its section says to.

check (read-only)

The batch hook (${CLAUDE_PLUGIN_ROOT}/hooks/bash-format.sh), the pre-commit check (${CLAUDE_PLUGIN_ROOT}/git-hooks/shell-check.sh) and the shared rewrite helper (${CLAUDE_PLUGIN_ROOT}/git-hooks/shfmt-rewrite.sh) are the single source of truth for what they require and how they resolve things.

Read it first. Probe what it actually does, don't recite this file. Then read the pre-computed tool rows, run the remaining probes via Bash, and report a PASS/FAIL/INFO table with one remediation line per FAIL. Do not modify anything. Read each option value from this text; never put a ${user_config.*} string in a shell command. Open plugin files by the ${CLAUDE_PLUGIN_ROOT} paths this skill names; never Glob or Grep the plugin directory.

The format pass (the batch row and the pre-commit rewrite) and the lint pass (the pre-commit ShellCheck findings) are independent; report each separately.

When the plugin's toggle is disabled, every prerequisite absence except node downgrades from FAIL to INFO. The hook exits through its enabled-gate before probing anything, so a deliberately disabled plugin is not broken. Report the probes informationally and note that re-enabling restores the FAIL semantics. A missing node stays FAIL in either state: every hook row launches node, and the gate runs inside hooks/exec-bash.mjs. A tool the git pre-commit entry needs keeps the grade it gets with the toggle on while that entry is installed (step 12 reports it as a tool line beside an ok, stale or modified entry): the entry runs at commit outside Claude Code and reads no toggle.

  1. Bash version. Check against the hooks' floor, Bash 3.2 or newer (${CLAUDE_PLUGIN_ROOT}/hooks/hook-utils.sh).
  2. node. The pre-computed node row. FAIL if absent: every hook row launches through node hooks/exec-bash.mjs, so without node the hook does not launch and no shell edit is formatted in the session (${CLAUDE_PLUGIN_ROOT}/README.md Requirements).
  3. jq. The pre-computed jq row. FAIL if absent: the batch row does nothing, since it reads the file paths with jq, and the pre-commit rewrite is skipped.
  4. shellcheck (lint pass). The pre-computed shellcheck row. FAIL if absent: the pre-commit entry reports no findings.
  5. shfmt (format pass). The pre-computed shfmt row. Its FAIL/INFO status depends on the .editorconfig opt-in below, because the format pass runs only when the repo has opted in:
    • opted in AND shfmt absent → FAIL: neither the batch row nor the pre-commit entry formats.
    • not opted in → INFO regardless of shfmt: the format pass stays quiet by design (the repo chose not to format), so a missing shfmt is not a defect here.
  6. .editorconfig shell opt-in. Mirror the hook's opt-in logic (shfmt_governed / section_applies_to_shell in ${CLAUDE_PLUGIN_ROOT}/git-hooks/shfmt-rewrite.sh), not merely "does an .editorconfig exist". The opt-in is an EditorConfig section that names shell files: a shell glob such as [*.sh], [*.bash], or [*.{sh,bash}] (including path-prefixed forms like [**/*.sh]), discovered by walking up from the file's directory to the repo root and stopping at a root = true config. A bare [*] catch-all does NOT count (most repos only set line-ending / charset properties there). Path-only sections like [scripts/**] do NOT count either. Report as INFO: whether a governing shell section exists and therefore whether the format pass is active. If none exists, INFO-note the consequence per the hook's logic: shell files are left unformatted rather than rewritten to shfmt's built-in defaults.
  7. .shellcheckrc. INFO: ShellCheck auto-discovers .shellcheckrc by walking up from the file's directory. Report whether one exists; its absence is not a FAIL (ShellCheck applies its own defaults).
  8. Hook options. Report the effective values. ${user_config.bash_format_enabled} (unexpanded or empty means the manifest default true; any value other than true disables the hook). ${user_config.bash_format_lint_gitignored} (unexpanded or empty means the manifest default false): unless it is true, the batch row silently skips an edit to a file the repository gitignores, so a user asking why .work/scratch.sh was not formatted gets this answer. ${user_config.bash_format_end_of_turn} (unexpanded or empty means the shipped default false), as INFO: on, the Stop row checks the uncommitted changes once per turn; off, only the batch format and the git pre-commit entry run.
  9. Hook registration. INFO: confirm the plugin is enabled for this project (/plugin → Installed) rather than parsing settings files.
  10. Project scope. INFO: when CLAUDE_PROJECT_DIR is set, the batch row acts only on shell files inside it (symlink-resolved membership guard in the shared hook library, aware of Windows 8.3 short-name spellings of in-project paths, a per-volume property; only volumes with 8.3 generation enabled produce them); a .sh/.bash file written outside the project (temp/scratchpad dirs) is silently skipped. No lint, no format, no notice. The OS temp tree counts as outside even when it sits under CLAUDE_PROJECT_DIR, unless the project root itself lives under temp. When CLAUDE_PROJECT_DIR is unset (e.g. some headless -p sessions) the guard is skipped and any existing edited file is processed. Report this so a green check is not read as "every shell edit anywhere is covered".
  11. Gitignored files. INFO: unless bash_format_lint_gitignored is true, a file the repository gitignores (git check-ignore) is silently skipped: no lint, no format, no notice. A tracked file that matches an ignore pattern stays in scope, and when git cannot decide (git absent, no repository, an error) the hook acts as before (hook::gitignored_out_of_scope in ${CLAUDE_PLUGIN_ROOT}/hooks/hook-utils.sh). Report the item 8 value beside this.
  12. Git pre-commit entry. Run node "${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs" verify from the repository. Each line is <state>\t<entry>[\t<detail>]: ok is PASS, declined is INFO (this clone said no; silent at session start), and missing, stale, modified or superseded is a FAIL whose remediation is apply install-git-hooks; a superseded detail lists what an install under an earlier plugin name left, which that action removes on the user's yes. A tool\t<tool>\tmissing line means the installed entry cannot run here: list every such tool with its install method from the ${CLAUDE_PLUGIN_ROOT}/prerequisites.json (Read that path), so a lane or cloud setup script can provision it before an auto-mode session commits. Outside a git repository, report INFO and skip the step.

When every probe passes, report the result with the scope caveats (items 10 and 11). Never an unqualified "fully operational", which would imply out-of-project shell edits are covered when they are deliberately skipped.

apply install-git-hooks [scope=local|shared]

Registers the plugin's git-hook entries (git-hooks/entries.json) with the repository's hook manager. The installer script only detects, renders and writes files the plugin owns; every step that needs judgment comes back to you as a decisions[] item, which you reason about and apply with your own edit tools.

  1. Scope. Use the argument when given. Otherwise, when the user is present, ask: local (this clone only, nothing committed) or shared (committed under .githooks/bash-format/, so every teammate and agent gets it). When nobody can answer, use local.
  2. Plan. Run node "${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs" plan --scope <scope> and read the JSON: manager, actions[], decisions[].
  3. Manager conflict. A manager-conflict decision lists the evidence (two managers, or one this installer does not drive, such as the pre-commit framework). Show it and ask which manager should carry the entry; rerun plan with --manager lefthook|husky|plain. Without an answer, stop here and report.
  4. Actions. When actions[] holds copy items, run node "${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs" copy --scope <scope> (plus the same --manager when you passed one). It prints installed, refreshed, unchanged, refused, blocked, unsafe-path or stage-failed per entry; refused means a locally edited copy or dispatcher, which the modified decision covers, blocked that another tool owns the local hooks directory, which the hooks-path-owned decision covers, unsafe-path the decision of that name, and stage-failed that git add failed on a shared install: report it, since the team would not receive the files.
  5. Decisions. Read reference/git-hook-decisions.md for what each kind asks of you, apply each one, then rerun plan until only the decisions you chose to leave remain.
  6. Verify. Run verify and report its lines as check step 12 does.

The entry rewrites: at commit it formats the staged shell scripts an .editorconfig section governs and stages the result, so tell the user that a commit can change the files it contains, and that a partly staged governed script is refused untouched.

When the user does not want an entry in this clone, run node "${CLAUDE_PLUGIN_ROOT}/lib/git-hooks/install.mjs" decline --entry shell-check; the session-start drift line then stays silent until verify runs by hand, and --undo reverses it. Add --global (with or without --undo) to decline the entry in every clone; git config --global checks.git-hooks.declined true declines every entry of every plugin, and an entry's own key set to false in a clone opts that entry back in. The adoption path explains to the user each scope and why a formatter's cleanup PR comes first.

apply (idempotent)

Run check, then for each FAIL point at the resolution. Apart from the git-hook entry above, this skill installs nothing:

  • missing shellcheck: install guidance from the README Requirements section (the ShellCheck install guide); this skill never installs system packages.

  • missing shfmt while the repo opts in: install guidance (shfmt); this skill never installs system packages.

  • missing node / jq / Bash: platform install instructions from the README Requirements section.

  • no shell .editorconfig opt-in (and formatting is wanted): explain that adding a governing shell section ([*.sh], [*.bash], or [*.{sh,bash}]) to an .editorconfig opts the repo in. A bare [*] is not enough, but this skill does not write it. .editorconfig is cross-cutting (it governs every editor and tool in the repo), so the choice and the edit belong to the consumer.

  • toggle off, gitignored files wanted, or the end-of-turn check wanted: set bash_format_enabled, bash_format_lint_gitignored or bash_format_end_of_turn to true through Claude Code's native flow, interactively with /plugin configure bash-format@<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 or pluginConfigs. Afterwards rerun check in a fresh session, since a same-session check can still show the old value, and report the observed effective value, never an unobserved change.

After pointing at a remediation, re-run the relevant check probe live via Bash (a pre-computed row predates the remediation, so never re-read it) and report its actual result. Never claim resolved on the reader's report that they installed something.

Re-running apply after everything passes changes nothing and reports "already configured".

What this skill does NOT do

  • Run the linter or formatter. Editing any .sh or .bash file exercises the batch row end to end; git commit exercises the installed pre-commit entry.
  • Write the plugin cache, Claude Code user settings, or pluginConfigs. Nor the repository, beyond the git-hook entry copies, including .editorconfig / .shellcheckrc. Every prerequisite is a PATH binary or the native toggle, so remediation is guidance only.
  • Download or execute tools during check beyond the read-only command -v presence probes.

Gotchas

  • ENAMETOOLONG when grepping the plugin cache. Installed plugins run from a deeply nested, cache-isolated path. Piping a grep/rg over the long absolute path to hooks/hook-utils.sh (or another bundled file) can spawn-fail with ENAMETOOLONG on some hosts. Read the file directly rather than passing the full cache path on the command line.
  • check PASS ≠ every shell edit is covered. When CLAUDE_PROJECT_DIR is set the hook is project-scoped (probe 10): shell files written outside it are silently skipped, so a fully green check still does not cover out-of-project edits. (When CLAUDE_PROJECT_DIR is unset the scoping does not apply. See probe 10.)
  • shfmt FAIL is opt-in-conditional. A missing shfmt is only a FAIL when an .editorconfig section governs shell files; without that opt-in it is INFO, not a defect. Resolve the .editorconfig opt-in state before calling shfmt a failure.

04 同仓库更多 Skills

melodic-software
melodic-software/claude-code-plugins

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.

★ 30
melodic-software
melodic-software/claude-code-plugins

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.

★ 30
melodic-software
melodic-software/claude-code-plugins

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.

★ 30
melodic-software
melodic-software/claude-code-plugins

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.

★ 30
melodic-software
melodic-software/claude-code-plugins

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.

★ 30
melodic-software
melodic-software/claude-code-plugins

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').

★ 30