agentleFS
Sn4kyGitSn4kyGit/qa-agent-blueprint/.claude/skills/skill-review/SKILL.md

skill-review

Skill2 starsChanged 8 days agoApache-2.0View on GitHub

Review which blueprint skills the evidence actually shows as used and propose a slimmer skill profile — read-only analysis of evidence/**/execution_summary.json (used/unknown with coverage, never "unused"), profile dry-runs (full/standard/minimal/custom) with dependency closure, apply only after explicit confirmation and only in integrate mode. Use for "/skill-review", "which skills do we need", "welche Skills werden genutzt", "skill profile", "skills reduzieren", "prune skills". Never reads transcripts and never deactivates a skill automatically.

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

3 rules. 0 are common, 2 are distinctive, and 1 is 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.

  • 1

    Very specific

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

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

    Read transcripts, chat logs, JIRA content, or screenshots.

    line 16
  2. Distinctive:

    Evidence gate: When the tracking holds no records at all (zero instrumented runs in the window), name no deactivation candidates and …

    line 68
  3. Distinctive:

    Before showing any proposal, close the keep-set over these rules and name every skill each rule pinned:

    line 101

“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 — evidence-based skill usage review

Two jobs, in order. First say what the evidence shows about skill usage. Only if the user then wants a slimmer skill set, propose a profile — as a dry run — and apply it only on an explicit confirmation and only where the install mode allows it. A proposal is never a decision; the user makes the decision.

What this skill never does

  • Read transcripts, chat logs, JIRA content, or screenshots. The only input is evidence/**/execution_summary.json.
  • Conclude "unused". Absence of a skills_used declaration is absence of instrumentation, not absence of use — the review reports unknown with a reason instead.
  • Deactivate, delete, or move anything automatically. Profile proposals are dry-run output; apply happens only per step 4.
  • Edit install.sh, opencode.json, plugin caches, or the blueprint checkout — and nothing at all outside integrate mode.

Step 1 — run the helper (read-only)

python3 .claude/scripts/skill-review.py [--evidence DIR] [--since DAYS] [--project KEY] [--json]
  • --evidence — evidence root, default evidence.
  • --since — window in days, default 90; 0 means no window.
  • --project — restrict to one project key.
  • --json — structured output for post-processing; the default render is for the user.

Python standard library only, writes nothing, reads execution_summary.json files and .claude/skills/*/SKILL.md names only. If .claude/skills/ is not found, the helper still reports declared usage and coverage and says the inventory is unavailable.

Step 2 — explain the result

Report three blocks, in this order:

  1. Coverage — the window, runs in window, runs carrying skills_used, the percentage, and separately: runs with unusable timestamps (kept and counted as timestamp_unknown, never silently dropped) and unreadable summaries.
  2. Used — per skill: count, last use date, source=declared_summary. This is the only claim the data supports: the run's own summary declared the skill.
  3. Unknown — every other installed skill with its reason: no_runs, field_not_instrumented, outside_test_run_scope, timestamp_unknown, unreadable_summary, retention_or_other_qa_home.

Say the rule out loud: unknown is not unused. skills_used is declared per run at summary time and only by flows that write a summary (.claude/rules/core/evidence-policy.md), so a missing declaration never proves a skill was not used. Old summaries without the field stay valid evidence — a coverage below 100 % mostly means runs predate the field, not that skills went unrecorded. Repeat runs or a wider window before drawing any conclusion.

Step 3 — profile proposal (only when asked)

Offer profiles only when the user asks to slim down, prune, or choose a skill set. Otherwise stop after step 2.

Evidence gate. When the tracking holds no records at all (zero instrumented runs in the window), name no deactivation candidates and propose no profile change: with no data, nothing can be named — say so, and at most point at the profiles below as a user preference decision that needs confirmation. "Which can I deactivate?" on zero evidence gets that honest answer, not a list.

Profiles

ProfileKeeps
full (default)every installed skill — the shipped state
standardall skills except agentswitch model-setup whats-new re-board demo scaffold-suite, then closed under the resolver
minimalstandard minus recon deploy-to-stage onboard-product onboard-project exploratory-testing, then closed under the resolver
customthe user's confirmed allowlist, closed under the resolver
  • full — everything. Nothing to propose, nothing to do.
  • standard — drops the harness-management and demo skills: agentswitch, model-setup, whats-new, re-board, demo, scaffold-suite. The ticket core, vendored specialists, recon, deploy-to-stage and skill-review itself stay.
  • minimal — standard minus recon, deploy-to-stage, onboard-product, onboard-project and exploratory-testing. The ticket core and skill-review itself stay.
  • custom — whatever the user names, then closed under the resolver below. A user's ad-hoc pick goes through the same resolver and the same confirmation as a named profile.

Profiles nest: minimal ⊂ standard ⊂ full. The vendored specialist skills and webdriverio-testing are pinned by the resolver when active rows or the framework require them, so no profile silently strands an active feature.

Dependency resolver

Before showing any proposal, close the keep-set over these rules and name every skill each rule pinned:

  1. Chain — test-ticket → failure-analysis → report-to-jira → retro-learn: keeping one link keeps the next. The same chain backs retest-bug, test-automation and exploratory-testing, so while any of them stays, failure-analysis, report-to-jira and retro-learn stay.
  2. qa-report and qa-help stay pinned in every profile — qa-report because the run-history scripts are invoked by path (failure-analysis/tc-history.py is called by failure-analyst; the paired history reader aggregate.py lives in qa-report), and qa-help because it is the command discovery surface every profile needs.
  3. webdriverio-testing only when projects/<KEY>/PROJECT.md → "Test assets" → Framework is webdriverio — keep it in exactly that case, drop it otherwise.
  4. Active feature-matrix rows pin their specialists — for every row with Applies: yes in the routed product's (or the project's) "Feature & test-method matrix", the skill named under "Responsible skill / subagent" stays: security-testing, api-testing, accessibility-testing, visual-testing. A profile that drops one strands an active row.

Without a project context, rules 3–4 cannot be evaluated: show the proposal without them and flag it as such. Never guess the framework or the matrix.

Show the proposal — dry run

Name the profile and the exact exclude list it would carry, one line per excluded skill with its reason (profile membership, resolver pin with its rule, or observed usage from step 2). State explicitly that nothing has been changed. Do not offer "apply" as a leading question — wait for the user to ask.

Step 4 — apply (explicit confirmation, integrate mode only)

Applying changes the consumer's skill inventory. All four conditions below must hold, or stop and explain which one does not.

  1. Integrate mode. Apply is only defined when the consumer owns real skill files: .claude/skills/ is a plain directory of the consumer repo, not a symlink into a blueprint checkout or plugin install, and the consumer is not the blueprint checkout itself. In plugin, opencode and shared mode the skill set ships whole — the plugin and the symlinked opencode surface are blueprint-owned and replaced on update, so nothing is ever deactivated there (docs/integration.md → "Skill profiles"). Explain that limitation and stop: the review stays read-only in those modes, and the same request gets the analysis plus the unapplied proposal.
  2. Explicit confirmation. Show the concrete proposal again (profile name + exact exclude list) and ask for a yes. No default, no carry-over: "mach es schlank" is not a confirmation — only an explicit yes to that concrete list is. If the list changes, show it again and confirm again.
  3. Apply through the installer's profile-apply step, never by hand:

    install.sh --mode integrate --target <consumer> \
      --skill-profile <name> --skills-exclude "<resolved exclude list>"

    The choice persists in the consumer's .qa-skill-profile.json (consumer-owned, backed up on every change) and is re-applied on every update/reinstall — a later update cannot silently bring the excluded skills back, and the installer re-resolves the dependency pins at apply time. Never rename or delete skill files yourself: that would desync the persisted profile and be undone by the next update. Never touch plugin caches or the blueprint checkout. If this checkout's installer does not implement profile-apply yet, stop and say so.

  4. Report what changed — the applied profile, the exclude list, and how to reverse it (--skill-profile full restores the complete set).

Excluded skills drop out of the helper's inventory afterwards (it lists */SKILL.md only) — expected; --skill-profile full restores both the skills and their appearance in the review.

Command surface

/skill-review — usage review only (steps 1–2). /skill-review <profile> or "reduce my skills" — steps 1–4, step 4 gated as above.

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.