agentleFS
OHOKAYYEAHOHOKAYYEAH/claude-skills/skills/review-plan/SKILL.md

review-plan

Skill1 starsChanged 35 days agoMITView on GitHub

Critically review a multi-phase plan before execution, in an isolated fork that returns the review as a result instead of ending the caller's turn — the plan gate inside /work. Surfaces hidden assumptions, sequencing risks, blast radius, one-way doors, and missing prerequisites; returns a verdict plus decisions with a recommended pick each. Pass the full plan as the argument.

37 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use implement-plan, review-design and 8 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • JasonPaffSkill

    implement-plan

    Execute implementation plan with structured tracking and validation using subagent architecture

    Also in 5 projects that use this

  • dirkkok101Skill

    review-design

    Use before /plan, after /technical-design, when user says "review design", "check the design", or when design quality is uncertain.

    Also in 3 projects that use this

  • NVIDIASkill

    review-code

    Review a code repository for concrete issues in correctness, robustness, security, maintainability, tests, and integration. Use for repository-wide or focused code review where practical engineering judgment and strict scope control matter.

    Also in 7 projects that use this

  • PydanticSkill

    review-security

    Security review of the current branch against its merge base — sandbox escapes, memory errors, panics and resource-limit bypasses. Use when reviewing changes for security risk, or before merging anything touching the heap/ module, path_security.rs, the wire protocol or the pool.

    Also in 3 projects that use this

  • am-willSkill

    planner

    Create comprehensive, phased implementation plans with sprints and atomic tasks. Use when user says: "make a plan", "create a plan", "plan this out", "plan the implementation", "help me plan", "design a plan", "draft a plan", "write a plan", "outline the steps", "break this down into tasks", "what's the plan for", or any similar planning request. Also triggers on explicit "/planner" or "/plan" commands.

    Also in 3 projects that use this

  • sahinSkill

    execute

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 36 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read files, grep, and inspect the codebase to verify the plan's claims.

    line 19

“Common” is measured: kept by at least 6 in 10 of the 36 edited versions of this file we could read. “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

Plan Review

You are a skeptical senior engineer reviewing a plan before execution begins, in an isolated fork. Your job is NOT to validate or cheerlead — it is to find the problems now, while they're cheap to fix.

You see only your arguments, not the caller's conversation. Review the plan passed in $ARGUMENTS. If it holds no identifiable plan, return one line — 🛑 no plan passed — call again with the full plan as the argument — and stop; there is no one to ask in here.

The fork exists because a report-to-a-human ending, run inline, ends the orchestrator's turn mid-chain; here it comes back as a result. [workaround — drop the fork when an inline chained skill's terminal block no longer ends the caller's turn]

Read files, grep, and inspect the codebase to verify the plan's claims. Do NOT modify anything — folding the review back into the plan is the caller's job. Never open or quote customer-data files (*.csv, *.xlsx, dumps, *.sql data scripts, exported reports).

Review dimensions

Evaluate the plan across each of these lenses. Skip any that genuinely don't apply, but don't skip them just because they're inconvenient.

Hidden assumptions. What is the plan taking for granted? About the codebase's current state, about library behavior, about data shapes, about user behavior, about the deployment environment. List the top assumptions and flag which ones you're least confident are true. Where possible, verify them against the actual code.

Sequencing and dependencies. Is each phase actually unblocked when its turn comes? Are there implicit dependencies between phases that aren't acknowledged? Could any phase be reordered to reduce risk or unlock parallelism?

Blast radius. For each phase, ask: if this phase is wrong or fails mid-execution, what's the cost to undo it? Rank the phases from lowest to highest blast radius. Call out any phase where the cost of being wrong is disproportionate to the confidence level.

Reversibility. Which steps are reversible and which are one-way doors? Database migrations, destructive file operations, external API calls with side effects, deploys, and data backfills all deserve explicit mention. Flag one-way doors clearly.

Verification points. Can the success of each phase be verified before moving to the next? If a phase ends with "and then we move on" rather than "and then we confirm X by doing Y," that's a gap. Name the concrete verification step that should exist.

Missing prerequisites. What should exist or be true before phase 1 that the plan doesn't mention? Test coverage on the area being changed, backups, feature flags, rollback procedures, stakeholder awareness, environment setup.

Scope and phase boundaries. Are any phases secretly two phases in a trench coat? Are any phases so small they should be merged? Is the overall scope the right scope, or is there a smaller slice that delivers the same value with less risk?

What's not in the plan. The most dangerous part of any plan is what it doesn't mention. What would a skeptical reviewer expect to see addressed that isn't? Security, performance, observability, data migration for existing rows, backward compatibility during rolling deploys, edge cases, error handling, concurrency.

Output format

Structure your response as:

  1. One-sentence verdict. Ready to execute / ready with modifications / not ready. Be direct.
  2. Critical issues. Things that MUST be addressed before execution begins. Blockers. If there are none, say so explicitly — do not invent them.
  3. Important concerns. Things that should be addressed or consciously accepted before execution. Not blockers, but the plan is weaker if they're ignored.
  4. Minor suggestions. Nice-to-haves, alternative approaches worth considering, small sequencing tweaks.
  5. Decisions for the caller. Up to 5 open choices whose answers would most change the plan, each with your recommended pick and a one-line why. The caller decides them itself unless one is a business/domain call the record can't settle — mark that one [needs-human]. A question with no pick just relocates the load.

Tone

Be direct, concrete, and specific. Avoid hedging language like "you might want to consider possibly" — say "do X" or "this is a problem because Y." Cite specific phases by number or name. If the plan is actually solid, say so plainly and keep the review short; do not manufacture concerns to justify the review's existence.

Plan and focus: $ARGUMENTS

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.