agentleFS
3orionkings3orionkings/forge/skills/forge-review/SKILL.md

forge-review

Skill0 starsChanged 19 days agoMITView on GitHub

Use when the user already has a plan or clear proposal and wants an adversarial cross-model review of that plan without a requirements interview or product build. Especially useful for auth, schema, concurrency, migrations, payments, and other high-cost decisions. Do not use for reviewing already-written code or trivial changes.

12 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use forge-ship.

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.
  • redteamlifeSkill

    forge-ship

    Use when the user asks to merge, release, promote, reconcile, or close out work in a repo containing `docs/forge/` or governed by FORGE. Not for per-task commits; those complete inside forge-execute-task.

    Also in 5 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    Plan: Do real planning — read the relevant code, surface decisions and tradeoffs — and write PLAN_FILE with: Goal · Approach · Key …

    line 20
  2. Very specific:

    Don't review already-written code — that's /codex:review.

    line 60
  3. Distinctive:

    Don't let the reviewer edit files, ever.

    line 61
  4. Distinctive:

    Don't skip the log — the argument transcript is the deliverable.

    line 62
  5. Very specific:

    Don't reconstruct codex exec commands from memory.

    line 63

“Common” is measured: kept by at least 6 in 10 of the 9 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

FORGE-REVIEW — Adversarial Plan Review, Standalone

Act 2 on its own: you already have a plan (or a clear idea) and want the cross-model stress test without the grill. Claude drafts/loads the plan; a different model attacks it read-only and returns VERDICT: APPROVED or REVISE; Claude revises and re-submits to the SAME session until APPROVED or MAX_ROUNDS.

Reach for it on high-stakes work — auth, schema, concurrency, migrations, payments. Skip it for obvious or cheap changes.

Unique to this skill

  1. Kickoff. Confirm scope in one line. If no task was given, ask for it (one question). Then proceed — no round-by-round approvals.
  2. Plan. Do real planning — read the relevant code, surface decisions and tradeoffs — and write PLAN_FILE with: Goal · Approach · Key decisions & tradeoffs (name the contestable choices so the reviewer has something to bite) · Risks / open questions · Out of scope. Initialize LOG_FILE.
  3. The loop. First perform the engine/auth/git preflight in ../forge/ENGINE.md, then open Round 1 in a fresh read-only session (../forge/references/invocation-templates.md §4a) and captures THREAD_ID; rounds 2..MAX resume that same session (§4b) so the reviewer never re-litigates settled points. Read this round's own verdict file — a per-run-unique path, never a shared fixed name — and confirm its mtime postdates the round's launch before trusting it.
  4. Each round: append the full critique to LOG_FILE; on REVISE, Claude decides what is worth acting on (Claude is final arbiter — the reviewer advises), revises the plan, and appends what changed and what was rejected with reasons.
  5. Resolution. On APPROVED, present the plan, a 3-bullet summary of what the argument improved, and the round count. On deadlock at MAX_ROUNDS, do not fake convergence — list each unresolved point with Claude's counter-position and hand it to the owner to break the tie.

Tunables: MAX_ROUNDS=5 · PLAN_FILE=PLAN.md · LOG_FILE=PLAN-REVIEW-LOG.md.

The reviewer is read-only every round — -s read-only on the fresh call, -c sandbox_mode="read-only" on every resume. That is a safety property, not cadence; it never relaxes. No product code is written in this skill, at either profile.

Handoff after the verdict

This skill ends at the plan verdict. If the user then asks to build, hand off to ../forge/SKILL.md and its selected execution profile. Do not load or restate Act-3/Act-4 cadence inside a plan-only review.

Canonical references

../forge/ROUTING-POLICY.md (routing) · ../forge/profiles/ (cadence) · ../forge/ENGINE.md (engine, templates, admission) · ../forge/references/ (detail, on demand)

What NOT to do

  • Don't review already-written code — that's /codex:review.
  • Don't let the reviewer edit files, ever.
  • Don't skip the log — the argument transcript is the deliverable.
  • Don't reconstruct codex exec commands from memory.

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.