菜单
Enforces manifest well-formedness when `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `.mcp.json`, or `.lsp.json` is opened for authoring or review — path-substitution rules, the hooks.json separation, and the house version policy.
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。
基于 SOC 职业分类
| name | plugin-manifest |
| description | Enforces manifest well-formedness when `.claude-plugin/plugin.json`, `.claude-plugin/marketplace.json`, `.mcp.json`, or `.lsp.json` is opened for authoring or review — path-substitution rules, the hooks.json separation, and the house version policy. |
| user-invocable | false |
| paths | ["**/.claude-plugin/plugin.json","**/.claude-plugin/marketplace.json","**/.mcp.json","**/.lsp.json"] |
Apply this to the file you just opened. The mistake that matters most: hand-bumping version in this repo — versions here are CI-managed, not author-managed. The second: registering hooks inside plugin.json instead of hooks/hooks.json.
name and description present; description is a single sentence stating what the plugin does.${CLAUDE_PLUGIN_ROOT} — in mcpServers/lspServers command/args/cwd/env, in hooks.json commands, in any bundled-script reference. Bare relative paths break once the plugin runs from a cache install rather than a checkout.hooks/hooks.json, not plugin.json. Keeping them separate lets /reload-plugins pick up hook edits without restarting MCP/LSP servers keyed off plugin.json.mcpServers/lspServers entries resolve their own command. For npm-shim binaries (.cmd/.bat, or node_modules/.bin shims) invoke node directly with the script path in args, not the shim by exec form.version field is NOT hand-edited in this repo. See house version policy below.marketplace.json entry source is either a relative path string ("./plugins/<name>") or an object form ({"source": "github"|"url"|"git-subdir"|"npm", ...}) — this repo uses both; don't assume relative-path is the only valid form. Any version set on the entry is superseded by plugin.json's version if both are present — don't set it in two places expecting independent control.claude plugin validate treats a misspelled or leftover field as a warning, not an error, unless --strict is used — always run with --strict here (see below)..mcp.json/.lsp.json at plugin root, not nested, unless the manifest's inline mcpServers/lspServers key is used instead — pick one form, don't duplicate a server in both.commands/agents/outputStyles/experimental.* REPLACE their default directory when set in plugin.json. The classic mistake: setting "commands": [...] and silently losing the default commands/ directory — list it explicitly to keep it (["./commands/", "./extras/"]). skills is the exception: it ADDS to the default skills/ scan rather than replacing it.**/.mcp.json and **/.lsp.json also match non-plugin project files. Confirm the file is at a plugin root (sibling of .claude-plugin/) before applying this checklist to it.This repo's plugins are versioned by CI. Do not hand-bump version in plugin.json or marketplace.json. If a change needs a version bump, that's a release-process concern, not something to fix inline while editing the manifest for an unrelated reason. Flag a manual version edit in review rather than making one yourself.
Manifest work is not done until this passes:
claude plugin validate <plugin-path> --strict
--strict promotes warnings (misspelled fields, wrong types, leftover fields from another tool's manifest) to errors — the default (non-strict) mode lets a plugin with only warnings pass and load anyway, which hides the kind of mistake this checklist exists to catch.
plugin.json instead of hooks/hooks.json.mcpServers/lspServers command — breaks under cache-install cwd.version bump alongside an unrelated manifest edit.claude plugin validate without --strict and treating a clean run as sufficient..cmd/.bat shim referenced via exec form (args present) — spawn fails; invoke the underlying script with node instead, or drop to shell form.${CLAUDE_PLUGIN_ROOT}/skills/anthropic-docs/references/plugins-reference.md — full manifest schema, component locations, path-substitution rules, ${CLAUDE_PLUGIN_ROOT}/${CLAUDE_PLUGIN_DATA} contracts, version resolution order, claude plugin CLI.${CLAUDE_PLUGIN_ROOT}/skills/anthropic-docs/references/mcp.md — .mcp.json server shapes, plugin-scoped tool naming, OAuth, timeouts.${CLAUDE_PLUGIN_ROOT}/skills/anthropic-docs/references/plugin-marketplaces.md — marketplace.json schema, plugin source variants, hosting, team/managed configuration.Scaffolding a brand-new plugin rather than auditing a manifest? Invoke the plugin-setup skill for the bootstrap checklist and the house hooks/lib/ templates.
Runs a dogfood cycle against plugin-bot — tasks the plugin-engineer agent with building a plugin capability inside the plugins/dogfood sandbox, validates and reloads it, evaluates the result (skill-creator evals for skills, fixtures/BATS for hooks), and harvests rough edges in plugin-bot's own skills and agent as improvement notes. Use after changing plugin-bot to exercise it end to end.
Enforces subagent frontmatter and system-prompt conventions when a `<name>.md` agent file is opened for authoring or review — plugin-scope ignored fields, description quality, tool-restriction shape, skills preload, and the required boundaries section.
Distilled, verification-stamped reference for the official Claude Code platform contract — plugins, skills, subagents, hooks, MCP, tools, env vars and channels. Use when authoring or auditing any Claude Code plugin component, looking up a frontmatter field, hook event schema, manifest rule, permission-rule format or tool name, or whenever a platform claim needs verifying instead of recalling from memory.
Enforces the house hook-script contract when a `hooks/**/*.sh`, `hooks/**/*.bash`, or `hooks/hooks.json` file is opened for authoring or review — envelope/exit-code/JSON discipline, subdirectory-per-event layout, lib helpers, fixtures, and BATS coverage.
Use when authoring, auditing, or registering a Claude Code plugin monitor — a monitors/monitors.json entry or the background watch script it runs. Ships the typed TypeScript poll-monitor harness (self-scheduling loop, stable-streak debounce, notify-once dedup, --once mode) plus the registration rules and notification-copy doctrine; auto-loads as the enforcer when a monitors/ file is opened.
Use when writing or reviewing language meant to shape an agent's behavior — a hook's `additionalContext` payload, a SKILL.md body or description, an agent system prompt, or an enforcer checklist. Covers choosing the imperative force, structuring with XML tags, matching the urgency tier to the moment (the EXTREMELY_IMPORTANT / IMPORTANT / reminder / acceptable-anti-pattern gradient), the persuasion-principles foundation (Cialdini + Meincke), why XML beats markdown for hook injections, and concrete templates for SessionStart, PostToolUse, and PreToolUse nudges.