agentleFS
MesteriisMesteriis/Engineering-Bible-AI/skills/multi-agent-pr-review/SKILL.md

multi-agent-pr-review

Skill0 starsChanged 4 months agoMITView on GitHub

Runs read-only parallel review of a PR, branch, commit, or diff across bugs, security, tests, maintainability, performance, and UI.

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read repository instructions and relevant changed files.

    line 14
  2. Distinctive:

    If subagent tools are available, dispatch independent read-only reviewers: security, runtime bugs, tests/flakiness, …

    line 15
  3. Distinctive:

    If subagents are unavailable, run the same categories sequentially.

    line 18
  4. Distinctive:

    Default: keep the current selected model.

    line 21
  5. Very specific:

    Use the task-scoped /multimodel and /quorum commands separately as specified in Explicit Activation.

    line 24
  6. Very specific:

    When /multimodel or /quorum is active, bind the same source snapshot, reviewed artifacts and acceptance criteria before dispatch.

    line 29
  7. Very specific:

    Only /quorum enables the fixed roster and voting decision policy in Worker Quorum: two-reviewer consensus or a three-reviewer quorum.

    line 35
  8. Distinctive:

    Do not edit files.

    line 64
  9. Distinctive:

    Do not report style-only issues unless they hide a real risk.

    line 65
  10. Very specific:

    Use security-router for deep security workflow when security findings are material.

    line 67

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Multi-Agent PR Review

Review a Git-backed change set with bounded read-only reviewers.

Workflow

  1. Resolve the exact base and head, or state that the current working tree is the target.
  2. Read repository instructions and relevant changed files.
  3. If subagent tools are available, dispatch independent read-only reviewers: security, runtime bugs, tests/flakiness, maintainability/architecture, performance, and UI/UX only when frontend files changed.
  4. If subagents are unavailable, run the same categories sequentially.
  5. Merge results with subagent-result-merge.

Default: keep the current selected model; multimodel and quorum are off. The skill name and parallel reviewers do not activate different models or votes. Same-model parallel reviewers remain available with the parent's selected model. Use the task-scoped /multimodel and /quorum commands separately as specified in Explicit Activation.

Cross-Provider Review

When /multimodel or /quorum is active, bind the same source snapshot, reviewed artifacts and acceptance criteria before dispatch. Use the existing worker result contract and a verified launcher; native subagent availability does not establish an external provider route. Host-specific mappings and credentials remain private. Apply agent-squad export and execution limits.

Only /quorum enables the fixed roster and voting decision policy in Worker Quorum: two-reviewer consensus or a three-reviewer quorum. Reviewers draft independently before seeing peers' results. Use distinct observed providers, exclude the change author from votes, and keep unknown routes, abstentions and failed launches out of approvals. An additional review is useful for material disagreement; small changes do not need a model panel. Do not repeat all project checks in every reviewer.

Without /quorum, merge findings without votes or majority thresholds. With /quorum, evaluate records using scripts/worker-evidence.py quorum and verify the actual source and artifact roots. Confirmed critical counterexamples and required failed checks block acceptance regardless of the approval count. The helper validates evidence consistency; it does not authenticate a provider, execute checks, establish isolation or authorize merging. The parent owns reproduction of findings, integration validation and applying changes.

Reviewer Contract

Each reviewer returns:

  • finding
  • severity
  • evidence with file/line
  • confidence
  • suggested minimal fix
  • validation gaps

Rules

  • Do not edit files.
  • Do not report style-only issues unless they hide a real risk.
  • Mark unverified concerns as uncertainty.
  • Use security-router for deep security workflow when security findings are material.

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.

Reviews can't be read right now.

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.