Plugin Review

以 Claude Code 执行者视角审查 plugin 产物（skills, agents, hooks, commands），发现逻辑 bug、触发机制问题和执行可行性缺陷。

DEPRECATED: This skill has been absorbed into skill-master:master. Use /master with review intent instead. This skill remains functional but will not receive updates.

Overview

This skill orchestrates plugin review across 9 dimensions. It detects whether the optional plugin-dev plugin is installed and routes accordingly:

• Strategy A (plugin-dev available): dispatches plugin-dev:plugin-validator + plugin-dev:skill-reviewer + own plugin-reviewer in parallel, then merges results
• Strategy B (plugin-dev not installed): dispatches own plugin-reviewer with supporting files to cover all 9 dimensions

Process

Step 1: Determine Scope

从用户消息判断审查范围：

Scope A — 指定文件：用户给出具体 skill/agent 文件路径。

Scope B — 指定 plugin：用户指定 plugin 名称或目录。收集该 plugin 下所有 skills 和 agents：

• Glob {plugin}/skills/*/SKILL.md → all skills
• Glob {plugin}/agents/*.md → all agents

Scope C — 最近变更：用户说"review my changes"或无明确指定。用 git diff --name-only HEAD 找到最近修改的 skill/agent 文件。如果 working tree 是 clean 的（已 commit），改用用户指定的范围（commit hash、HEAD~N、--since 等），或默认 git log --name-only -1 --pretty=format:"" 获取最后一次 commit 的变更文件。

Scope D — 全量审查：用户说"review all"或"audit everything"。收集所有已安装 plugin 的 skills 和 agents。

在审查范围确定后，列出将被审查的文件清单，让用户确认。

Step 2: Gather Context

对审查范围内的每个 plugin，收集：

1. Plugin manifest — .claude-plugin/plugin.json
2. All skill files — skills/*/SKILL.md
3. All agent files — agents/*.md
4. Marketplace registration — 检查 .claude-plugin/marketplace.json 是否包含该 plugin
5. Eval files — 对每个 skill 目录，检查是否存在 skills/{name}/eval.md

构建文件清单传递给 agent。

Step 2.5: Detect plugin-dev Availability

Check whether the plugin-dev plugin is installed:

• Glob for ~/.claude/plugins/*/plugin-dev/.claude-plugin/plugin.json or .claude/plugins/cache/*/plugin-dev/*/.claude-plugin/plugin.json
• If found: use Strategy A
• If not found: use Strategy B

Record the strategy choice for Step 4.

Step 3: Dispatch (Strategy A or B)

Strategy A — plugin-dev available (3 parallel dispatches in a single message):

1. plugin-dev:plugin-validator agent:

Validate this Claude Code plugin's structure.
Plugin manifest: {path}
Skills: {comma-separated paths}
Agents: {comma-separated paths}

2. plugin-dev:skill-reviewer agent:

Review these skill descriptions for trigger quality and routing clarity.
Skills: {comma-separated paths}

3. skill-audit:plugin-reviewer agent with model: "opus":

Review these Claude Code plugin artifacts from the AI executor perspective.

Scope: {A: specific files | B: plugin | C: recent changes | D: all}
Files to review:
- Skills: {comma-separated paths}
- Agents: {comma-separated paths}
- Plugin manifest: {path}

Also read these for cross-reference checking:
- Other skills in same plugin(s): {paths, for trigger conflict detection}
- Other agents in same plugin(s): {paths, for reference integrity}
- Eval files: {comma-separated paths or "none"} — for trigger plausibility checking

Supporting files to load: none
Plugin agents dir: !`echo "${CLAUDE_PLUGIN_ROOT}/agents"`
(D1/D2 structural checks and baseline trigger/description checks are handled by plugin-dev agents.)

Focus on: workflow logic, execution feasibility, edge cases, dispatch loops, spec compliance, metadata & docs, eval.md consumption, deep trigger conflict detection, and Trigger Health Score.
Do NOT review code style or formatting — only functional correctness.

Strategy B — plugin-dev not available (1 dispatch):

skill-audit:plugin-reviewer agent with model: "opus":

Review these Claude Code plugin artifacts from the AI executor perspective.

Scope: {A: specific files | B: plugin | C: recent changes | D: all}
Files to review:
- Skills: {comma-separated paths}
- Agents: {comma-separated paths}
- Plugin manifest: {path}

Also read these for cross-reference checking:
- Other skills in same plugin(s): {paths, for trigger conflict detection}
- Other agents in same plugin(s): {paths, for reference integrity}
- Eval files: {comma-separated paths or "none"} — for trigger plausibility checking

Supporting files to load: structural-validation.md, trigger-baseline.md
Plugin agents dir: !`echo "${CLAUDE_PLUGIN_ROOT}/agents"`

Focus on: logic bugs, trigger mechanism issues, execution feasibility, and edge cases.
Do NOT review code style or formatting — only functional correctness.

Step 4: Present Results

When all dispatched agents complete:

Strategy A — merge results:

1. If any agent dispatch failed or timed out: mark its dimensions as "Error: agent did not return results" in the Dimension Summary and proceed with available results
2. Collect all findings from the 3 agents, tagging each finding with its source agent
3. Deduplicate: if multiple agents flag the same issue (same file + same location), keep the more specific finding
4. Map findings to the 9-dimension Dimension Summary table. Use finding content to determine the appropriate dimension row — do not rely on dimension IDs from external agents' output format
5. Assemble into the unified Plugin Review Report format (same structure as Strategy B output)

Strategy B — direct presentation: Present the plugin-reviewer output directly.

Both strategies:

1. Group findings by severity:
   • Bug — will cause incorrect behavior or execution failure
   • Logic — won't fail but reduces effectiveness or produces misleading results
   • Minor — cosmetic or low-impact concerns
2. For each Bug-severity finding, include the suggested fix inline
3. If fixes exist: ask user "Apply suggested fixes?" and apply if approved

Completion Criteria

• Review report presented with findings grouped by severity
• Every finding has a file:line reference and specific description
• Bug-severity findings include actionable fix suggestions
• If user approved fixes: changes applied