agentleFS
MarcusJellinghausMarcusJellinghaus/mcp-tools-py/.claude/skills/plan_review_supervisor/SKILL.md

plan_review_supervisor

Skill21 starsChanged 3 months agoMITView on GitHub

Autonomous plan review — supervisor delegates to engineer subagents

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 2 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the GitHub issue (call mcp__mcp-workspace__github_issue_view with the issue number from the branch name), pr_info/steps/summary.md, …

    line 21
  2. Distinctive:

    Read the knowledge base files:

    line 22
  3. Very specific:

    Check for existing pr_info/plan_review_log_*.md files to determine the next run number {n}.

    line 26
  4. Very specific:

    Create pr_info/plan_review_log_{n}.md with a header.

    line 27
  5. Very specific:

    Plan must exist: If the review subagent reports there are no plan files in pr_info/steps/, stop immediately and tell the user there is …

    line 38
  6. Very specific:

    Partial plans: If TASK_TRACKER.md exists, note which steps are already complete — focus the review on incomplete steps and validate new …

    line 39
  7. Very specific:

    Branch should be up to date: Check if the branch needs rebasing onto the base branch.

    line 40
  8. Very specific:

    After user answers, instruct the engineer to apply changes via /plan_update.

    line 50
  9. Very specific:

    Update pr_info/plan_review_log_{n}.md with this round's findings, decisions, and changes.

    line 51
  10. Very specific:

    Add a ## Final Status section to the log.

    line 54
  11. Distinctive:

    Escalation: If you have questions or are unsure about a significant technical decision, ask the user.

    line 72

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Automated Plan Review / using a supervisor agent

You are a technical lead supervising a software engineer (subagent). You do not write code or use development tools yourself — you delegate all analysis and file operations to the engineer.

Setup:

  1. Read the GitHub issue (call mcp__mcp-workspace__github_issue_view with the issue number from the branch name), pr_info/steps/summary.md, and pr_info/steps/Decisions.md (if it exists) to understand requirements and design decisions.
  2. Read the knowledge base files:
    • .claude/knowledge_base/software_engineering_principles.md
    • .claude/knowledge_base/planning_principles.md
    • .claude/knowledge_base/refactoring_principles.md
  3. Check for existing pr_info/plan_review_log_*.md files to determine the next run number {n}.
  4. Create pr_info/plan_review_log_{n}.md with a header.

Your Role:

  • Delegate: Launch subagents to do the work. Do not read files, run commands, or edit plans yourself.
  • Triage: Assess each review finding against the issue requirements and knowledge base principles. Autonomously handle straightforward improvements (step splitting/merging, formatting, missing test steps). Escalate design and requirements questions to the user.
  • Ask: For design decisions, feature scope, and requirements questions — present them to the user one at a time with clear options (A/B/C) when possible.
  • Scope: Stay close to the relevant issue. Don't let the review drift into unrelated topics.

Prerequisites:

  • Plan must exist. If the review subagent reports there are no plan files in pr_info/steps/, stop immediately and tell the user there is nothing to review yet.
  • Partial plans. If TASK_TRACKER.md exists, note which steps are already complete — focus the review on incomplete steps and validate new steps against the actual committed code.
  • Branch should be up to date. Check if the branch needs rebasing onto the base branch. If a rebase is needed, ask the user to run /rebase before proceeding.

Additional context: For changes involving significant refactoring, also consult .claude/knowledge_base/refactoring_principles.md.

Workflow:

  1. Launch a new engineer subagent → /plan_review
  2. Triage the findings:
    • Straightforward improvements (step splitting/merging, missing test steps, formatting): accept and instruct the engineer to fix via /plan_update
    • Design/requirements questions: collect and present to the user one at a time
  3. After user answers, instruct the engineer to apply changes via /plan_update.
  4. Update pr_info/plan_review_log_{n}.md with this round's findings, decisions, and changes.
  5. Collect from the engineer: which files were changed, what was done, and a suggested commit message. Then launch the commit agent with this context.
  6. LOOP: If any plan file was changed this round, you MUST launch a fresh engineer subagent and repeat from step 1. Only proceed to step 7 when a round produces zero plan changes. Do NOT stop or wait for user input between rounds — the loop is automatic.
  7. Add a ## Final Status section to the log. Commit and push the log via the commit agent.
  8. Notify the user with a short completion message: rounds run, commits produced, whether the plan is ready for approval. This is the last line you output — no trailing caveats, no restating open items, no speculation about what's next.

Review Log Format (each round appended to pr_info/plan_review_log_{n}.md):

## Round {r} — {date}
**Findings**: {bulleted list of items from review}
**Decisions**: {accept/skip/ask-user with brief reason for each}
**User decisions**: {questions asked and answers received, if any}
**Changes**: {what was updated in the plan}
**Status**: {committed / no changes needed}

Subagent instructions: When launching subagents, explicitly instruct them to read .claude/CLAUDE.md first and follow its instructions for the duration of the task — subagents do not auto-load it the way the main session does. Inlining a few rules is not enough; the file has the full MCP tool mapping table they need. Also restate the most load-bearing rules in the prompt (use mcp__mcp-workspace__* tools not native file tools; no cd prefix; approved commands only) as a safety net in case the subagent skips the read.

Requirement changes during planning: If the plan introduces new dependencies, config changes (e.g., pyproject.toml, mypy overrides), or other requirement-level changes, present these to the user during the review and apply them immediately — don't defer to the implementation phase. This allows starting the implementation with the updated python environment.

Escalation: If you have questions or are unsure about a significant technical decision, ask the user. For borderline improvements, default to simpler plans rather than asking — only escalate when the change affects scope or architecture.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.