agentleFS
samuelralaksamuelralak/How-I-Work/skills/adversarial-review/SKILL.md

adversarial-review

Skill0 starsChanged 5 days agoMITView on GitHub

Brief, carry out or act on an independent adversarial review. Use before committing a slice, to recheck a fix round, when you are briefed as the reviewer, and when the owner asks for a drift review of architecture, design or standards.

105 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use rethink, root-cause-analysis and 10 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.
  • fabioespindulaSkill

    rethink

    Reframe a plan, idea, or decision before the user commits to it. Use when the user asks to rethink, step back, pressure-test, gut-check, decide, right-size, simplify, think bigger about, or evaluate a concrete plan or decision in product, strategy, technical architecture, career, life, company direction, or personal priorities. This skill improves decision quality before action: clarify the real question, challenge assumptions, compare alternatives, calibrate ambition, classify the operating mode, surface risks, and recommend a next move. Do not use for broad future exploration without a concrete decision; that belongs to a think-big or research-style skill.

    Also in 3 projects that use this

  • MicrosoftSkill

    root-cause-analysis

    Use when a Regional Retail Manager asks I'm seeing unusual inventory movement at our Northwest stores. Can you help me understand what's happening?

    Also in 3 projects that use this

  • anasahmed07Skill

    better-auth-best-practices

    Skill for integrating Better Auth - the comprehensive TypeScript authentication framework.

    Also in 4 projects that use this

  • DockerSkill

    fix-issue

    Fix a single GitHub issue end-to-end: triage, research, write the fix, review, and create a PR. Use when asked to fix an issue: "fix issue 1234", "resolve #500", "create a PR for issue 200".

    Also in 3 projects that use this

  • MicrosoftSkill

    pull-request

    Write concise, focused pull request descriptions for Wassette — at most three sentences on the what and why, with an explicit note on any breaking public-API change and how users adapt. Use when opening or editing a Wassette pull request.

    Also in 3 projects that use this

  • juriwiensSkill

    codex-review

    Run code review via Codex CLI

    Also in 3 projects that use this

  • +2 more from microsoft

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 94 edited versions we read keep these.

  • 2

    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:

    No praise, and no summary of what the slice does.

    line 56
  2. Distinctive:

    Ask for: a one-paragraph verdict, findings by severity with the rule quoted, systemic patterns with counts, documents that are out of date, …

    line 76

“Common” is measured: kept by at least 6 in 10 of the 94 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

Adversarial review

One independent reviewer checks a slice before it is committed. Its job is to find what is wrong, not to confirm that the work is done. If you are the reviewer, "Rules" and "The report" are yours; the rest is for whoever briefs you.

Rules

  • The reviewer is a separate agent that did not build the slice. It never edits the code under review. If you cannot start one, write the brief to a file and ask the owner to open a fresh session with it.
  • Give it the code and the claims, never your conclusion. Ask it to attack; do not ask it to approve.
  • It reviews the exact change that will be committed: the diff of the slice's paths against the head, or the builder's patch when the commit is built apart from the tree.
  • It may run single targeted tests, never the full suite, a paid call or anything against shared data. Two agents never run tests on one database at once: it uses the builder's only after the builder's runs have stopped, or else one of its own, dropped when it reports.
  • A finding is a defect traced in the code, probed with a small test when it rests on what happens at run time. What it could not trace is listed apart as a question.
  • Two passes: the review, then one recheck of what changed. After that only a blocker still standing earns another.

The brief

Tell the reviewer to load this skill; the brief then carries only what is specific to the slice:

  1. What the slice is, in two or three sentences, and where the diff and the builder's notes are.
  2. The sources of truth: the standards file, the design documents, earlier reviews of the same area.
  3. What the slice claims to do, as a list. These are claims to test.
  4. What to look hardest at. Draw from this list:
    • correctness under retries, restarts, races and two tabs;
    • safety: outside text reaching a model, private data reaching a log or a broadcast, access to another person's records, output escaping;
    • money: who is charged, how often, and what bounds it;
    • data: schema-only changes, existing rows, anything needing a step after deploy;
    • what changes for live behaviour the moment it deploys;
    • tests, against the agreement's Tests rules: does each name read as a promise; take the fix away, does it fail for that reason; does it assert an outcome; any sleep, test id, styling class, wall clock or real outside call; and which promise in the change has no test;
    • organisation: files in their fixed places, names, duplication, dead code, drift from the agreed design;
    • the standards file, rule by rule, on everything the slice touches.
  5. Which test database it may use, and what never to run.

A small slice (a few files; no money, safety or data) needs items 1 and 3, the organisation check and a read of its tests only. The last review of a feature of several commits also reads the whole feature for drift.

The report

  • Findings, most severe first. For each: severity (blocker, should fix, minor), file and line, one sentence on the defect, and the concrete scenario that goes wrong.
  • What was checked and found sound, briefly.
  • One closing line: commit as is, commit after the blockers, or do not commit.
  • No praise, and no summary of what the slice does.

After the report

  1. Save the report with the slice's other records.
  2. Decide what must be fixed before the commit: every blocker, and every finding of the same kind as a blocker (safety, money, data loss, a wrong answer). Product choices go to the owner with a recommendation.
  3. The builder reads the whole report, fixes in one pass, each fix with a test that fails first, and answers every finding in one line. You settle a disputed finding by reading the code.
  4. The same reviewer rechecks what changed. Nothing is committed while a blocker stands; everything else goes on the written follow-up list, with file and line.
  5. A reviewer can be wrong. Before a finding reaches the owner or changes a plan, open the code or the data it rests on yourself.

Drift review

For a review of a whole codebase against its standards, when the owner asks for one:

  • One read-only reviewer per area, for example backend and interface, on the committed code. The standards file is law; drift the convention tests do not catch is the most valuable to find.
  • Ask for: a one-paragraph verdict, findings by severity with the rule quoted, systemic patterns with counts, documents that are out of date, and what was not covered. Harm to a person, money or data is kept apart from tidiness.
  • Bring the owner a ranked summary and let them choose what to fix first. Fix in reviewed slices, each with a guard so the rule stays fixed.

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.