brutal-pr-review
Skill2 starsChanged 4 months agoMITView on GitHub
Review an open pull request with Brutal multi-perspective scrutiny, snapshot-safe BRUTAL.md context, and provider-native findings. Supports normal, strict all-findings, and material-convergence passes.
In this repository25 other files it gives its agents.
- brutal-deepresearchSkill
- brutal-grillSkill
- brutal-idea-evalSkill
- brutal-inf-fix-loopSkill
- brutal-planSkill
- brutal-pr-finding-fixerSkill
- brutal-pr-fix-loopSkill
- brutal-project-reviewSkill
- brutal-realitySkill
- brutal-reviewSkill
- brutal-specSkill
- brutal-swarmSkill
- brutal-ticketsSkill
- brutal-wayfinderSkill
- brutal-workerSkill
- codex-coopSkill
- design-jepa-modelsSkill
- factory-designSkill
- funcvec-dedupe-fixerSkill
- grill-meSkill
- handoffSkill
- opencode-delegateSkill
- rust-architecture-reviewSkill
- rust-pr-reviewSkill
- supply-chain-analyzerSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
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
- Very specific: line 18
Read applicable repository rules and only the docs/manifests needed to interpret the changed surface.
- Distinctive: line 23
Use provider comments only.
- Distinctive: line 25
Stop before posting if either the base SHA or head SHA changes after context collection.
- Very specific: line 38
When native reviewer subagents are available, launch them with fork_turns: "none" and give only the exact review snapshot.
- Very specific: line 55
Every result includes validated_finding_count, material_finding_count, counts by severity and placement, queued count, residual MINOR/NIT …
- Distinctive: line 77
Before posting, re-read base/head SHAs.
≈ “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.
Files
Brutal PR Review
Review the exact open pull request resolved through BRUTAL.md.
Required Context
- Read
../brutal-shared/integration-resolver.mdand../brutal-shared/support/contracts.md. - Reuse validated integrations, canonical local root, and exact pull request from a managed phase. Otherwise resolve one code host and one work store and read both support modules.
- Require an open pull request. Never create one or substitute a local diff.
- Read applicable repository rules and only the docs/manifests needed to
interpret the changed surface. Load
references/rust.mdfor Rust changes.
Hard Rules
- Use provider comments only. Never approve, request changes, merge, or write to forks without explicit permission.
- Stop before posting if either the base SHA or head SHA changes after context collection.
- Validate every finding against the current diff and surrounding code.
- Patch only owned comments beginning with an exact canonical marker. Recognize legacy markers for queue compatibility but write only v2 markers.
Snapshot And Reviewer Isolation
Key a review context by provider/repository, pull-request ref, base branch, base SHA, head branch, and head SHA. A matching head alone is insufficient: a base-only change invalidates the context. Reuse verification only for that exact snapshot and only while required checks still pass.
When native reviewer subagents are available, launch them with
fork_turns: "none" and give only the exact review snapshot. They must not
inherit the worker’s implementation conversation. If subagents are
unavailable, apply the same evidence isolation sequentially. Merge only
validated structured findings.
Review Modes
normal: validate all severities; publish/queue CRITICAL and MAJOR. Report MINOR/NIT only in the summary.strict_all: publish/queue all four severities. Clean means zero validated findings.material_convergence: validate all four severities. If CRITICAL or MAJOR exists, publish/queue all four. If neither exists, publish no actionable MINOR/NIT comments; include those residuals in the summary and returnmaterially_clean. Zero findings returnszero_findings.
Every result includes validated_finding_count, material_finding_count,
counts by severity and placement, queued count, residual MINOR/NIT findings,
posting actions, checks, and completion kind when converged.
Review Records And Posting
Generate a safe unique review_id. Give each finding a deterministic
fingerprint and begin its body with:
<!-- brutal-pr-review:v2:<fingerprint> -->
<!-- brutal-pr-review-occurrence:<review_id>:<reviewed_head_sha> -->For each enabled severity, post line-mappable findings inline and other findings as individual top-level comments. Never hide an enabled actionable finding only in the summary.
For GitHub, send the final validated set to
scripts/post_github_review.py. Use inline_severities for the mode-derived
actionable severities: CRITICAL/MAJOR for normal or a converged material pass,
all four for strict mode or a material pass that found CRITICAL/MAJOR. Add
--dry-run only for an explicit preview or safety guard.
Before posting, re-read base/head SHAs. Return the review id, complete reviewed snapshot, finding counts, queue counts, summary action, checks, and blockers.
Redirect verbose output to temporary logs. Report command status and duration; on failure include at most the last 200 lines or 16 KiB, whichever is smaller.
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
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.

