agentleFS
rayketcham-labrayketcham-lab/qn-claude-web/.claude/skills/grill/SKILL.md

grill

Skill2 starsChanged 8 months agoApache-2.0View on GitHub

Adversarial code review — stress-tests recent changes by hunting for bugs, edge cases, security holes, and design flaws. Spawns parallel attack agents, consolidates into a severity-rated report.

92 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use write-skill, root-cause and 9 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.
  • NoaiRoxSkill

    write-skill

    Use when creating a new skill, editing an existing skill, or verifying a skill works before deployment.

    Also in 5 projects that use this

  • Ingramradical235Skill

    root-cause

    Structured root cause analysis using WHERE (MECE decomposition) then WHY (causal structure diagrams) pipeline. Includes coin-flip detection, logical-leap detection, and vicious cycle intervention. Use when buffer enters YELLOW/RED, when metrics stall, or when Actions execute but don't produce results.

    Also in 4 projects that use this

  • sahinSkill

    execute

    Also in 5 projects that use this

  • assisjpSkill

    file-pr

    Prepare and open or update an authorized pull request with a concise problem-first description and verifiable evidence. Use when the user asks to create, file, open, or update a PR. Do not use merely to review a diff or when publication is not authorized.

    Also in 3 projects that use this

  • Arman-KudaibergenovSkill

    write-plan

    Этот скилл MUST быть вызван когда есть design.md и нужно разбить доработку на атомарные задачи (2-5 мин) с точными путями файлов и критериями проверки в tasks.md. SHOULD также вызывать как промежуточный шаг между brainstorm и subagent-dev. Do NOT использовать для обсуждения идеи — используй brainstorm; для выполнения задач — используй subagent-dev.

    Also in 3 projects that use this

  • JkudjoSkill

    incident

    Production incident response — structured triage, hypothesis tree, safe remediation, incident record

    Also in 3 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. 0 are common, 13 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 72 edited versions we read keep these.

  • 13

    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:

    If $ARGUMENTS is empty or "recent", target uncommitted changes (git diff + git diff --cached).

    line 18
  2. Very specific:

    If $ARGUMENTS is a number, treat it as a PR number and fetch the diff with gh pr diff $ARGUMENTS.

    line 19
  3. Very specific:

    If $ARGUMENTS is a file path, scope the review to that file and its direct dependents.

    line 20
  4. Distinctive:

    Read every changed file in full.

    line 42
  5. Distinctive:

    If there are tests, do they actually assert the right thing?

    line 72
  6. Distinctive:

    All three agents run in parallel.

    line 76
  7. Distinctive:

    After all agents return, merge their findings.

    line 80
  8. Distinctive:

    If Architect flags a design flaw AND SecOps finds an exploitable path through it, escalate severity.

    line 81
  9. Distinctive:

    If Tester identifies a missing test AND SecOps found an untested security boundary, link them.

    line 82
  10. Distinctive:

    If Architect sees coupling AND Tester sees that the coupled code has no integration test, flag it.

    line 83
  11. Distinctive:

    Only MEDIUM/LOW = PASS WITH CAVEATS or PASS

    line 138
  12. Distinctive:

    Before merging anything non-trivial

    line 146
  13. Distinctive:

    After a Builder implementation, before Verifier signs off

    line 147
  14. Distinctive:

    When reviewing a PR from an external contributor

    line 148
  15. Distinctive:

    When you suspect something is wrong but can't pin it down

    line 149
  16. Distinctive:

    After a production incident, reviewing the code that caused it

    line 150

“Common” is measured: kept by at least 6 in 10 of the 72 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

/grill -- Adversarial Code Review

Tear apart recent changes. Find what breaks, what leaks, what fails at 3 AM under load. No politeness, no benefit of the doubt. Every line is guilty until proven correct.

Target

$ARGUMENTS

If $ARGUMENTS is empty or "recent", target uncommitted changes (git diff + git diff --cached). If $ARGUMENTS is a number, treat it as a PR number and fetch the diff with gh pr diff $ARGUMENTS. If $ARGUMENTS is a file path, scope the review to that file and its direct dependents.

Process

Step 0 -- Gather the Diff

Determine the attack surface. Run the appropriate command based on the target:

# Uncommitted changes (default)
git diff
git diff --cached
git diff --stat

# PR number
gh pr diff $PR_NUMBER

# Specific file
git log --oneline -10 -- $FILE
git diff main -- $FILE

Read every changed file in full. Understand what the code does, what it replaced, and what it touches. Map the blast radius: which modules depend on the changed code? What calls it? What does it call?

Step 1 -- Parallel Attack Agents

Launch three adversarial agents concurrently using the Task tool. Each gets the full diff and surrounding context. Each has one job: find problems.

  1. Architect (subagent_type: architect, model: "opus") -- Think deeply about all design flaws. Attack design and structure:
    • Does the abstraction make sense, or does it leak?
    • Are there coupling violations, circular dependencies, or layering breaks?
    • Does the API contract make promises the implementation can't keep?
    • Are there race conditions, ordering assumptions, or shared-state hazards?
    • Would this design survive 10x scale? 100x?
    • Is there dead code, unreachable branches, or cargo-culted patterns?
    • Target: $ARGUMENTS
  2. SecOps (subagent_type: secops, model: "opus") -- Think deeply about all attack vectors. Enumerate every possible exploit. Attack security posture:
    • Input validation: what happens with malicious, malformed, or oversized input?
    • Injection vectors: SQL, command, path traversal, template injection, LDAP, header
    • Auth/authz: can this be called by the wrong principal? Is there a TOCTOU gap?
    • Crypto: weak algorithms, hardcoded keys, nonce reuse, timing leaks
    • Secrets: anything that looks like a credential, token, or key in the diff?
    • Dependency risk: new deps with known CVEs or excessive permissions?
    • Error disclosure: do error messages leak internal state to callers?
    • Target: $ARGUMENTS
  3. Tester (subagent_type: tester) -- Attack test coverage and correctness:
    • What inputs are NOT tested? Empty, null, max-length, negative, Unicode, concurrent
    • What error paths have no test? Timeout, OOM, disk full, permission denied, network down
    • What boundary conditions are missed? Off-by-one, overflow, underflow, wrap-around
    • Are there implicit assumptions that would fail on different OS, locale, or timezone?
    • If there are tests, do they actually assert the right thing? (Tautological tests are worse than no tests -- they give false confidence.)
    • What regression test is missing if this is a bug fix?
    • Target: $ARGUMENTS

All three agents run in parallel. Wait for all to complete before proceeding.

Step 2 -- Consolidation and Cross-Reference

After all agents return, merge their findings. Look for compound issues:

  • If Architect flags a design flaw AND SecOps finds an exploitable path through it, escalate severity.
  • If Tester identifies a missing test AND SecOps found an untested security boundary, link them.
  • If Architect sees coupling AND Tester sees that the coupled code has no integration test, flag it.

Deduplicate. If two agents found the same issue from different angles, combine into one finding with both perspectives.

Step 3 -- Rate and Report

Assign severity to each finding:

SeverityCriteria
CRITICALExploitable in production. Data loss, auth bypass, RCE, or crash under normal load. Fix before merge.
HIGHLikely to cause bugs or outages. Wrong under edge conditions that will occur in production. Fix before merge if possible.
MEDIUMCode smell, missing test, weak design that will cause pain later. Should fix, can merge with a tracked issue.
LOWStyle, minor improvement, defense-in-depth suggestion. Nice to have.

Output Format

# Grill Report: [Target]

**Scope**: [N files, M lines changed]
**Reviewed by**: Architect, SecOps, Tester (parallel)
**Verdict**: PASS / PASS WITH CAVEATS / FAIL

## Findings

### [CRITICAL] [Short title] -- [agent source]
**Location**: `file:line`
**What's wrong**: [Direct explanation. No hedging.]
**Trigger**: [How to exploit or reproduce this.]
**Fix**: [Concrete suggestion, not "consider improving."]

### [HIGH] [Short title] -- [agent source]
...

### [MEDIUM] [Short title] -- [agent source]
...

### [LOW] [Short title] -- [agent source]
...

## Summary
- Critical: N
- High: N
- Medium: N
- Low: N

## Verdict Rationale
[Why this passes or fails. If PASS WITH CAVEATS, list exactly what must be tracked.]

Verdict rules:

  • Any CRITICAL finding = FAIL
  • 2+ HIGH findings = FAIL
  • 1 HIGH finding = PASS WITH CAVEATS (must be tracked)
  • Only MEDIUM/LOW = PASS WITH CAVEATS or PASS

Tone

Be the staff engineer who has been paged at 2 AM because of code like this. You are not here to encourage. You are here to find what breaks. Every finding must be specific, actionable, and grounded in the actual code -- not generic advice. If the code is solid, say so briefly and move on. Do not manufacture findings to fill space.

When to Use

  • Before merging anything non-trivial
  • After a Builder implementation, before Verifier signs off
  • When reviewing a PR from an external contributor
  • When you suspect something is wrong but can't pin it down
  • After a production incident, reviewing the code that caused it
  • Any time the user invokes /grill

When NOT to Use

  • Documentation-only changes (unless docs describe security procedures)
  • CI/CD config tweaks with no code impact
  • Dependency version bumps with no API changes (use SecOps solo for CVE check instead)

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.