agentleFS
IT-HUSETIT-HUSET/andthen/.claude/skills/skill-review/SKILL.md

skill-review

Skill4 starsChanged 4 months agoMITView on GitHub

Review a skill bundle, a prompt-like file (`AGENTS.md`, a guideline, an agent definition), or a change to them for conflicts, text past its allowance, and dropped parsed contracts, returning findings, a branch ledger, and a ship-ready verdict; `--fix` rewrites from the ledger. Trigger on 'review this skill change', 'is this AGENTS.md lean', 'tighten this prompt'.

30 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use sigma, skill-forge and 2 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.
  • sanyuan0704Skill

    sigma

    Personalized 1-on-1 AI tutor using Bloom's 2-Sigma mastery learning. Guides users through any topic with Socratic questioning, adaptive pacing, and rich visual output (HTML dashboards, Excalidraw concept maps, generated images). Use when user wants to learn something, study a topic, understand a concept, requests tutoring, says 'teach me', 'I want to learn', 'explain X to me step by step', 'help me understand', or invokes /sigma. Triggers on: learn, study, teach, tutor, understand, master, explain step by step.

    Also in 3 projects that use this

  • WilliamSaysXSkill

    skill-forge

    Automated skill creation workshop with intelligent source detection, smart path management, and end-to-end workflow automation. This skill should be used when users want to create a new skill or convert external resources (GitHub repositories, online documentation, or local directories) into a skill. Automatically fetches, organizes, and packages skills with proactive cleanup management.

    Also in 3 projects that use this

  • GoogleSkill

    grill-me

    A relentless interview to sharpen a plan or design.

    Also in 4 projects that use this

  • SentrySkill

    skill-creator

    Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, writing, registration, and verification.

    Also in 5 projects that use this

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 24 edited versions we read keep these.

  • 2

    Distinctive

    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 the working tree as it stands, because that is what the next commit ships.

    line 14
  2. Very specific:

    After the review, read fix.md and follow it.

    line 18
  3. Very specific:

    Read SKILL-AUTHORING-GUIDELINES.md (the guideline) whole before the review.

    line 24
  4. Distinctive:

    Search the shipped surface and the project's docs for each rule the scope states, because a paraphrase of a rule stated elsewhere is …

    line 46
  5. Very specific:

    Run the Findings Filter as Devil's Advocate over the scope: does the span fail the rule, does the model do worse because of it, and is the …

    line 66
  6. Very specific:

    Gate: every loaded file is in the size ledger, the conflict audit has a result, every branch and aspect in scope has its row, every …

    line 79

“Common” is measured: kept by at least 6 in 10 of the 24 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

Skill Review

Reviews what a prompt's text makes the model do and what it costs. Frontmatter validity is the host validator's job.

Input

$ARGUMENTS minus flags, and the request around it, name the scope: a skill bundle, one or more prompt-like files (an instruction file, a guideline, an agent definition, an output style, a task-start document), or a change to them, such as a diff, a commit range, or the working tree's edits. Read the working tree as it stands, because that is what the next commit ships.

  • --fix rewrites the scope from the branch ledger, and a direct imperative in the request ("tighten", "rewrite") counts as --fix. Only it edits. After the review, read fix.md and follow it.
  • --output-dir <path> writes the report there as skill-review-<stem>.md, or skill-review-<stem>-fix.md under --fix. Without it the report prints inline.

Rules

  • Read SKILL-AUTHORING-GUIDELINES.md (the guideline) whole before the review. It is the rubric, and every finding cites its rule.
  • The loaded path is the files one load pays for: a bundle's SKILL.md, every reference it names, and agents/openai.yaml, or a file plus what it pulls in on the same turn.
  • Findings and the ledger cover the scope, judged in its loaded path. For a change, the scope is the changed text.
  • ../../../plugin/references/review-calibration.md owns the Anti-Leniency Protocol, Scope Discipline, the Structured Finding Contract, and the Findings Filter, so a remediation reader parses this report like any other. The Critic posture in ../../../plugin/references/lens-adversarial.md is the finding pass.
  • Collect the Project Rules Context per ../../../plugin/references/intent-and-rules-context.md, including the surfaces a parsed contract's reader lives in: README sections, tests, eval cases, Key Dev Commands.
  • The size ledger is wc -c per loaded file, their sum, and how often that text loads: a skill when it fires, an instruction file or guideline every turn, an agent definition every spawn, a task-start document nearly every run. It is the only size claim, so "too long" without its numbers is no finding.
  • A task-start document (LEARNINGS.md, DECISIONS.md) is data. Review it for Duplication, Sediment, and each entry's truth against the code and the current skills.

Review

1. Conflict audit

Audit conflicts across the loaded path, the project instruction file, and the host prompt.

2. Critic pass

Attack the scope against the guideline. Search the shipped surface and the project's docs for each rule the scope states, because a paraphrase of a rule stated elsewhere is Duplication however well each copy reads. Each finding names the span (file and line), the guideline rule it fails, and what the model does differently because of it.

3. Branch ledger

One row per branch and aspect in scope: the span, its aspect, its skeleton section on a bundle, and the verdict its allowance sets. A cut row carries the characters it frees and its keeper, or decision. The ledger is the rewrite's plan, and text past its allowance is a finding only where a change adds it, or it is a truism or a span the guideline's What no longer belongs catalogue names. A ledger without a cut on a body written before the current model generation is the reviewer under-firing, so re-walk it before reporting Ready.

4. Leading-word probe

Probe every cut to a leading word in one run of python3 scripts/cold-probe.py <prompt-file>, the prompt under .agent_temp/, carrying none of the target's text and asking per term recognised: yes | partly | no and the behaviour it implies. The script loads no instruction file, so the answer is not borrowed from rules that already define the term. A recognised term replaces the span, keeping what the project adds beyond the probe's answer; otherwise the span stays. Without the claude CLI, probe with a no-tools subagent (the installed worker role agent when available, else a generic inherited one). A term its loaded instructions or skill descriptions name cannot be probed cold there and keeps its definition.

5. Filter

Run the Findings Filter as Devil's Advocate over the scope: does the span fail the rule, does the model do worse because of it, and is the severity the one below? Report the Filter summary line and each withdrawal's falsifier.

Severity and readiness

  • CRITICAL – an instruction conflict, or a parsed contract dropped, broken, or contradicted: a dispatch naming a skill or argument that does not exist, a parsed token misspelt.
  • HIGH – text a change adds past its allowance, a trigger-surface defect, a phase without a gate, a procedure at a dispatch site.
  • MEDIUM – a rule paraphrased across files, a rule against the model's default without its reason, a truism or catalogue span no change adds, data in running prose, a step or rule whose three or more instructions run as one paragraph, a task-start document entry no longer true.
  • LOW – a paragraph carrying several concerns, a packed sentence, a source line holding more than one sentence, a leading word not reused, a missing table of contents, an unqualified tool name.

The readiness line is Blocked on any CRITICAL, else Needs Fixes on any HIGH or three or more MEDIUM, else Ready. It names the host validator the target still has to pass (claude plugin validate inside a Claude Code plugin, the loose-skill installer when the project ships one) or says none applies.

Gate: every loaded file is in the size ledger, the conflict audit has a result, every branch and aspect in scope has its row, every leading-word cut carries its probe result, and every surviving finding carries Class: and Routing:. A change to a parsed contract, or a move that changes a registration, routes Note with its fix written as SURFACED: <what> – <what would resolve it>.

Output

Verdict first, per the calibration: the readiness line, the size ledger with the characters the cuts free, the conflict audit, the findings in the contract's shape, the branch ledger, the Filter summary, and Critic Coverage. Under --fix, add every item of the gate in fix.md.

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.