agentleFS
sturlesesturlese/build-squad/skills/fix/SKILL.md

fix

Skill0 starsChanged 3 months agoMITView on GitHub

Bug-fix pipeline, reproduce-first — the tester captures the bug as a failing test BEFORE the developer may touch production code, then verifies the fix and scans for sibling occurrences of the same defect. Accepts an informal report or a screenshot and clarifies interactively in the main session first; when the bug genuinely cannot be reproduced up front, falls back to writing the regression test right after the fix. Use when the user reports a bug, regression, or broken behavior and wants the team to fix it.

82 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use feature, frontend-code-review and 8 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.
  • bradtraversySkill

    feature

    Manage current feature workflow - start, review, explain or complete

    Also in 10 projects that use this

  • chendongqiSkill

    frontend-code-review

    Trigger when the user requests a review of frontend files (e.g., `.tsx`, `.ts`, `.js`). Support both pending-change reviews and focused file reviews while applying the checklist rules.

    Also in 4 projects that use this

  • GoogleSkill

    pr-creator

    Also in 4 projects that use this

  • PJuniszewskiSkill

    cook

    Feature development with guardrails. Plan → Review → Code → Ship.

    Also in 3 projects that use this

  • vbrevikSkill

    work

    Use when the user says 'plan', 'todo', 'copy plan', 'append plan', 'resume plan', 'priorities', or 'what's next'.

    Also in 4 projects that use this

  • Microsoft AzureSkill

    test

    Also in 12 projects that use this

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 75 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    Context: Read CLAUDE.md and README.md if present.

    line 24
  2. Very specific:

    Docs — documentator, conditional: Only if user-visible behavior, contracts, or documented flows changed.

    line 34
  3. Distinctive:

    No production code before the brief is confirmed (informal intake) or the gate is passed.

    line 40
  4. Distinctive:

    Test runs are sequential unless the project explicitly documents its suites as parallel-safe.

    line 43

“Common” is measured: kept by at least 6 in 10 of the 75 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

Team fix

Bug report: $ARGUMENTS (plus any screenshots pasted in the conversation)

You are the orchestrator. One principle drives this pipeline: a bug that is not reproduced is a rumor, and a fix without a failing test is a cover-up. Reproduce-first is the rule — no production code changes until a red test exists — with a single explicit, human-gated exception (step 2).

Intake (main session, only when the report is informal)

If the bug arrived as a screenshot or a few loose sentences, close the brief HERE, in your own session, before spawning anyone — clarification is a conversation (a subagent relay would be slow and lossy) and subagents never see the conversation's images, so a screenshot's facts must be extracted by you into text first.

  • Ingest. Extract every fact from the report and screenshots: literal error messages, UI state, visible data, environment hints. Read the project's context (CLAUDE.md, README.md, overview command if documented).
  • Investigate before asking. Locate the affected area in the code and form a hypothesis. Never ask the user something the code, the logs, or the screenshot already answer.
  • Clarify in small rounds. Ask ONLY what still blocks the fix — 2–3 questions per round, each with your best-guess answer attached so the user can confirm or correct — then stop as soon as the brief closes. The brief is closed when it has: expected behavior (the contract), actual behavior, reproduction path, suspected area, and a "fixed when …" acceptance line. Show it compactly for one confirmation — a nod, not a negotiation. If the user prefers to fix later, save the brief to specs/bugs/<slug>.md and stop; this skill resumes from that file.

A precise report needs no intake — go straight to the choreography.

Choreography

  1. Context. Read CLAUDE.md and README.md if present; if the project documents a system-overview command (e.g. a make info target), run it. Pass the relevant slice to every agent you spawn.
  2. Reproduce — tester leads. Spawn the tester with the report/brief and context. Its mission: reproduce the bug and capture it as the smallest failing automated test (or, when automation is genuinely impossible, a precise manual reproduction procedure — flagged as such). It also diagnoses the affected contract: is this actually a bug (behavior diverges from the contract) or a contract-change request in disguise?
  3. GATE — human confirms. Present: reproduction evidence, the failing test (path and name), the affected contract, and the tester's diagnosis. Exits:
    • Confirmed bug, reproduced as a red test → continue, reproduce-first.
    • Confirmed bug, but genuinely not reproducible up front (needs live/prod state) → the explicit exception: proceed to the fix and write the regression test AFTER, as a receipt that would have failed before it. The gate must say this out loud; the human owns the call.
    • Intended behavior / feature request in disguise → stop and offer /squad:define to spec it (it closes quickly for small scopes) followed by /squad:build.
    • Nothing reproduced and no path forward → stop; report what was tried and exactly what information is missing.
  4. Fix — developer. Spawn with the failing-test reference (or the confirmed brief, in the exception), the tester's diagnosis, and a minimal-fix mandate: make the failing test pass WITHOUT weakening it, touching as little as possible. The developer must not modify the test (the tester owns it) — if the test itself seems wrong, it reports that instead of editing it. No opportunistic refactors: improvement ideas go in the report as follow-ups.
  5. Verify — tester (∥ auditor when sensitive). Tester re-runs the new test (must be green) plus the smallest regression set around the touched area, sequentially by default; in the exception, this is where the regression test is written and proven. Spawn the auditor in parallel ONLY if the diff touches sensitive territory (auth, data integrity, money, concurrency, security) — otherwise skip it and say so in the report. Route failures back to the developer with evidence; maximum 2 cycles, then escalate to the user.
  6. Sibling scan. Have the developer (or the auditor, if spawned) scan for the same defect elsewhere: the same helper misused at other call sites, a symmetric flow (create vs update) sharing the broken pattern, copy-pasted logic. Siblings are reported as follow-ups — never fixed silently in this run.
  7. Docs — documentator, conditional. Only if user-visible behavior, contracts, or documented flows changed; otherwise skip and say so in the report.
  8. Delivery report. Root cause in one paragraph; the failing-test-now-green proof; files changed; regression targets run and results; sibling occurrences found; docs touched; follow-ups.

Rules

  • Reproduce-first is the default and the whole point of the pipeline; the fix-then-test exception (step 2) is taken only at the human gate, only when up-front reproduction is genuinely impossible, and is named as such in the report.
  • No production code before the brief is confirmed (informal intake) or the gate is passed.
  • Minimal fix only; refactors and sibling fixes ride in follow-ups, not in bug fixes.
  • The developer must not weaken or modify the tester's regression test; disagreements are reported, not edited away.
  • Test runs are sequential unless the project explicitly documents its suites as parallel-safe.
  • Artifacts and reports in English; talk to the user in their own language.

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.