メニュー
Audit plugins for sub-agent opportunities — verbose skills, coverage gaps, over-permissions. Use when reviewing where sub-agents would help or auditing agents against the always-Opus standard.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/laurigates/claude-plugins --skill agents-analyzeコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SOC 職業分類に基づく
Install, configure and troubleshoot MCP servers. Use when adding/enabling servers, editing .mcp.json, fixing OAuth, or when a server runs stale code after an upstream fix.
FinOps snapshot — org billing, workflow stats, cache usage. Use when you want a high-level view of CI spending or workflow health before diving deeper.
GitHub Actions billing, workflow efficiency, and waste analysis at org or repo level. Use when investigating CI/CD costs, wasted runs, or optimizing triggers.
SKILL.md を表示中
| description | Audit plugins for sub-agent opportunities — verbose skills, coverage gaps, over-permissions. Use when reviewing where sub-agents would help or auditing agents against the always-Opus standard. |
| args | [--focus <plugin-name>] |
| allowed-tools | Glob, Grep, Read, Bash(ls *), Bash(wc *), TodoWrite |
| model | opus |
| argument-hint | analyze all plugins or --focus <plugin-name> |
| created | "2026-01-24T00:00:00.000Z" |
| modified | "2026-07-18T00:00:00.000Z" |
| compatibility | claude-code |
| reviewed | "2026-06-18T00:00:00.000Z" |
| name | agents-analyze |
| agent | general-purpose |
| context | fork |
Analyze the plugin collection to identify where sub-agents would improve workflows by isolating verbose output, enforcing constraints, or specializing behavior.
| Use this skill when... | Use a sibling skill instead when... |
|---|---|
| Auditing the whole plugin collection for sub-agent opportunities (verbose-output skills, model mismatches, tool over-permissions) | Auditing a single agent's frontmatter, tool list, and prompt completeness — use agent-patterns-plugin:meta-audit |
| Mapping delegation gaps and producing a list of proposed new agents | Authoring the new agent file from that proposal — use agent-patterns-plugin:custom-agent-definitions |
| Confirming every agent runs on Opus (the always-Opus standard) | Configuring an agent's hooks, permissions, or settings.json wiring — use hooks-plugin:hooks-configuration |
Focusing the analysis on a single plugin's skills (--focus <plugin>) | Coordinating multiple agents at runtime — use agent-patterns-plugin:agent-teams or parallel-agent-dispatch |
| Context | Command |
|---|---|
| List all plugins | find . -maxdepth 1 -type d -name '*-plugin' |
| Count skills per plugin | find <plugin>/skills -name 'SKILL.md' -o -name 'skill.md' | wc -l |
| List existing agents | find agents-plugin/agents -maxdepth 1 -name '*.md' |
| Check agent model field | grep -r '^model:' agents-plugin/agents/ |
| Check agent allowed-tools | grep -r '^allowed-tools:' agents-plugin/agents/ |
| Skill tool permissions | grep -r '^allowed-tools:' */skills/*/SKILL.md |
find . -maxdepth 1 -type d -name '*-plugin'find . -path '*/agents-plugin/agents/*' -maxdepth 3 -name '*.md'find . -path '*/skills/*/skill.md'find . -path '*/skills/*/SKILL.md' -not -path './agents-plugin/*'$1: Optional --focus <plugin-name> to analyze a single plugin in depthPerform a systematic analysis of the plugin collection to identify sub-agent opportunities.
Scan the repository to build an inventory:
agents-plugin/agents/ to understand current coverage--focus is provided, restrict analysis to that plugin onlyFor each plugin (or focused plugin), evaluate skills and commands against these criteria:
Operations that produce verbose output benefiting from isolation:
| Indicator | Examples |
|---|---|
| Build tools | docker build, cargo build, webpack, tsc |
| Infrastructure ops | terraform plan/apply, kubectl describe |
| Test runners | Full test suite output, coverage reports |
| Profiling tools | Flame graphs, benchmark results |
| Security scanners | Vulnerability reports, audit output |
| Log analysis | Application logs, system logs |
| Package managers | Dependency trees, audit results |
Operations that should be limited to specific tools:
| Constraint | Rationale |
|---|---|
| Read-only analysis | Security audit, code review - no writes |
| No network | Pure code analysis tasks |
| Limited bash | Tasks that shouldn't execute arbitrary commands |
Every plugin agent runs on model: opus. A subagent's output re-enters the main loop as a tool result, so a weaker delegate quietly degrades everything downstream — and Opus-low beats Sonnet-high on both quality and tokens. So effort (a session setting), not model, is the cost lever: a mechanical agent stays on Opus and dials effort down rather than downgrading the model.
| Audit finding | Recommendation |
|---|---|
Agent file declares model: opus | OK — no change |
Agent file declares sonnet / haiku / any non-opus | Flag it — recommend model: opus, and note that mechanical agents tune effort down instead |
Agent file omits model: | Flag it — agents require an explicit model: opus |
The sole sanctioned non-Opus subagent is the agent-patterns-plugin:cold-read-gate haiku reader, which is a skill-inline Agent(model: haiku) dispatch (the measurement instrument), not an agent file — so no */agents/*.md is exempt. The scripts/check-agent-model.sh lint enforces this; see .claude/rules/agent-development.md § "Model Selection for Agents".
Compare identified opportunities against existing agents:
model: opus (always-Opus standard)/delegate references agents that don't existFor each recommended new agent, specify:
### Proposed: <agent-name>
- **Model**: opus (always; tune `effort` down for mechanical agents, never the model)
- **Covers plugins**: <list>
- **Context value**: <what verbose output it isolates>
- **Tools**: <minimal set>
- **Constraint**: <read-only, no-network, etc.>
- **Priority**: HIGH | MEDIUM | LOW
- **Rationale**: <why this is better than inline execution>
For model/tool corrections to existing agents:
### Fix: <agent-name>
- **Current model**: <non-opus> → **Recommended**: opus (then dial `effort` down if mechanical)
- **Reason**: <why the change improves things — e.g. restores the always-Opus standard>
If new agents are recommended, check:
/delegate commandopus (always-Opus standard; effort is the cost lever, not the model)## Sub-Agent Analysis Report
**Scope**: [All plugins | focused plugin name]
**Date**: [today]
**Plugins analyzed**: N
**Existing agents**: N
**Skills without agent coverage**: N
### Current Coverage Map
| Domain | Agent | Skills Covered | Gaps |
|--------|-------|----------------|------|
| ... | ... | ... | ... |
### Recommended New Agents
[Proposals from Step 4]
### Recommended Fixes
[Model/tool corrections from Step 4]
### Delegation Mapping Updates
[Any updates needed for /delegate command's agent reference table]
### Priority Summary
| Priority | Count | Top Recommendation |
|----------|-------|-------------------|
| HIGH | N | ... |
| MEDIUM | N | ... |
| LOW | N | ... |
After presenting the analysis:
agents-plugin/agents/agents-plugin/README.md with new agents/delegate command's agent reference table if needed