agentleFS
TheAstroByteTheAstroByte/astrobyte-agent-skills/skills/astrobyte-review/SKILL.md

astrobyte-review

Skill1 starsChanged 14 days agoMITView on GitHub

Review a change against intent, repository standards, correctness, security, data integrity, maintainability, and real verification evidence. Use for PRs, branches, diffs, risky artifacts, or explicit adversarial review.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is 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.

  • 1

    Very specific

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

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

    Read repository standards and affected contracts.

    line 12
  2. Distinctive:

    Run a focused proof for any high-impact finding that can be checked cheaply.

    line 23
  3. Distinctive:

    Start the report with a decision layer that states:

    line 31
  4. Distinctive:

    Keep detailed findings and evidence below that layer or behind artifact links.

    line 39

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Astrobyte review

  1. Fix the comparison point and gather the complete diff or artifact set.
  2. State the intended outcome from the request, issue, spec, commit history, and code.
  3. Read repository standards and affected contracts. Resolve delivery and review preferences from this skill's scripts/runtime_config.py --project <project-root> and configuration.md. Missing local configuration uses the packaged review policy without onboarding. Apply current instructions and project evidence before personal defaults.
  4. Review these axes:
    • intent and acceptance criteria
    • correctness and edge cases
    • data integrity, provenance, and migrations
    • security and trust boundaries
    • public interfaces and compatibility
    • bloat. See principles.
    • tests and real-runtime evidence
    • documentation and operational recovery
    • PR text. Describe the change and its verification for repository reviewers. Remove internal workflow narration and unrelated setup advice while retaining material code risks and verification limits.
  5. Run a focused proof for any high-impact finding that can be checked cheaply.
  6. In durable and production modes, run the review gate before an agent-created pull request or push. In explore and prototype, run it when the operator asks for a review.
    • Start with the packaged review policy. Apply the returned working_agreement_layers in order, including their review instructions. Portable differences precede legacy local agreements; local deltas follow them. Apply higher-priority machine, project, and task preferences afterward. Resolve the reviewer against the models and independent read-only routes the current harness actually offers. Strongly recommend Sol and prefer it when available. If Sol is unavailable and no explicit reviewer requirement applies, ask the operator to select an available independent reviewer. An explicit requirement stays binding; when it cannot be met or no independent route exists, report the blocker and keep the gate pending. The push or pull request waits for a current independent PASS. Read reviewer-dispatch.md for dispatch. Missing configuration uses the public preference; self-review never satisfies this gate.
    • Send that reviewer one read-only strict brief: the intent, the comparison point, the diff, and any external feedback the change must satisfy. Ask for a first line of exactly PASS or BLOCK, followed by findings with file locations. Any other reply has not passed, and never infer a verdict from the wording. Ask the reviewer once for its verdict; if the second reply is also unclear, report the gate pending and the push or pull request waits.
    • On BLOCK, when this run is also implementing the change, fix the finding. When the finding conflicts with the spec, a cited pattern, or a human reviewer's instruction, send that evidence back and ask for a re-ruling. The reviewer stays read-only.
    • A fresh review must return PASS. Stop and report after three blocked rounds.
  7. Synthesize as lead reviewer. Deduplicate findings and account for missing context.

Start the report with a decision layer that states:

  • the practical problem;
  • the smallest recommended action;
  • why it works;
  • what remains risky or unproven;
  • the exact decision or action required from the operator.

Keep detailed findings and evidence below that layer or behind artifact links.

Classify each item:

  • act: blocks correctness, safety, data integrity, or the requested outcome
  • consider: real tradeoff that may be worth addressing
  • note: valid context with low present impact
  • dismiss: incorrect, unsupported, or out of scope

Lead with act findings ordered by severity and include file or artifact locations. State when no actionable findings remain. Do not apply changes unless the user also requested implementation.

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.