fix
Skill187 starsChanged 5 months agoMITView on GitHub
Use to fix a reported bug spec-first: reproduce it, trace the symptom to the owning feature's acceptance criteria, pin it with a failing (RED) test, apply the minimal GREEN fix through the same per-task gate implement uses, then patch the spec so the bug class can't silently return. Triggers on "fix {bug}", "fix the bug in {slug}", "bug in {feature}", "/sdd:fix {slug}", "regression in {slug}", "полагодь баг", "виправ багу", "регресія в {slug}", "чому зламалось". Triage is three-way: AC exists and is violated (regression) / AC is ambiguous (spec-bug — patch the wording) / no AC covers it (gap — add one, marked added-by-fix). Works on a repo with no specs at all (soft mode — code-first, recommends survey after). Writes a fix record under docs/features/{slug}/_fixes/ and commits the fix (SDD-Fix trailer), the record and any spec patch as separate commits.
82 other projects use a version of this.
Adopted by
Most recently taken up by
- tyrantpiper/travel-pwatheir own version, first seen Nov 8, 2026
- Eclipter/agentstheir own version, first seen Nov 8, 2026
- Bhenderlong/vallorix-mcp-plugintheir own version, first seen Nov 8, 2026
- KadirHarmanc/nazar-skilltheir own version, first seen Nov 8, 2026
- zedarvates/frugal-skillstheir own version, first seen Nov 8, 2026
- ribato22/vibechecktheir own version, first seen Nov 8, 2026
Used with
Projects with it also use feature, frontend-code-review and 9 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.
bradtraversySkillfeature
Manage current feature workflow - start, review, explain or complete
Also in 10 projects that use this
chendongqiSkillfrontend-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
GoogleSkillpr-creator
Also in 4 projects that use this
PJuniszewskiSkillcook
Feature development with guardrails. Plan → Review → Code → Ship.
Also in 3 projects that use this
vbrevikSkillwork
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 AzureSkilltest
Also in 12 projects that use this
Other versions82 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 82
- react/react★ 252ksame name
- facebook/react★ 252ksame name
- motiondivision/motion★ 35ksame name
- parcadei/Continuous-Claude★ 3.9ksame name
- parcadei/Continuous-Claude-v2★ 3.9ksame name
- parcadei/Continuous-Claude-v3★ 3.9ksame name
- crbnos/carbon★ 2.8ksame name
- Akylas/OSS-DocumentScanner★ 2.6ksame name
- kdlbs/kandev★ 1.1ksame name
- GoogleCloudPlatform/magic-modules★ 975same name
72 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository22 other files it gives its agents.
- apiSkill
- clarifySkill
- classify-sizeSkill
- configSkill
- data-modelSkill
- decide-adrSkill
- designSkill
- design-systemSkill
- glossarySkill
- implementSkill
- interviewSkill
- plan-testsSkill
- reviewSkill
- roadmapSkill
- scaffoldSkill
- screensSkill
- sequencesSkill
- shipSkill
- specifySkill
- surveySkill
- tasksSkill
- ux-flowsSkill
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 75 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
- Very specific: line 38
No depth dial and no .size here — a fix is one size, and the interview is the bug report itself.
- Distinctive: line 65
Patching code when the spec was the bug: If the AC permitted the behaviour, the wording is the root cause.
- Distinctive: line 66
Silent spec edits: Every AC patch/addition is confirmed with the user — the spec is a contract, not a scratchpad.
“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.
Files
Skill: fix
The bugfix entry point — the backbone in miniature, sized for «it's broken», not «build a feature». A bug is treated as evidence about the spec, not only about the code: either an acceptance criterion is violated (the code regressed), the AC was ambiguous enough to permit the behaviour (the spec is the root cause), or nothing covers it (a gap). So the fix always lands in two places — the code (RED → GREEN through the same gate implement runs) and the spec (a surgical AC patch) — tied together by a small fix record.
This skill keeps only its own machinery. Question phrasing → ../_shared/ask-style.md; RED-classification semantics → ../implement/references/tdd-loop.md (reused, never duplicated); dispatch policy → ../_shared/agent-roster.md.
Fix-record prose follows artifact_language — but a spec patch matches the existing spec's language (the file wins over the setting); code, tests and commits stay English → ../_shared/artifact-language.md.
Owner
The engineer on the bug (drives). PM / Tech Lead is consulted only when triage lands on «spec-bug» or «gap» — changing an AC is a product decision, not a code one.
Inputs
<slug>— optional; pass it when you know which feature owns the bug, otherwise step 2 finds it from the symptom.- The bug report, in any form — a sentence, a stack trace, a failing request, a screenshot description.
- Soft gate (never hard-refuse):
docs/features/with ≥1spec.md. Absent (a brownfield repo that never ran the backbone) → still run, in no-spec mode: steps 1 → 3 → 4 → record, skip the spec patch, and recommend/sdd:surveyin the handoff. - (Optional)
.claude/sdd.local.md— gate command overrides; otherwise the commands are detected perimplement's cascade.
No depth dial and no .size here — a fix is one size, and the interview is the bug report itself.
Protocol
- Intake — reproduce before touching anything. At most 1–2
AskUserQuestion(phrasing per../_shared/ask-style.md), only for what the report doesn't already say: expected vs actual, the steps, the scope (one user? all? since when?). Outcome: a one-line reproduction statement — «doing X, expected Y, got Z». A bug you can't state this way isn't ready to fix. - Trace to spec (triage). Grep
docs/features/*/spec.md(+ the candidate slug's_fixes/for a recurrence) for the reproduction's domain terms; locate the owning slug and the closest §5 AC. In parallel, dispatchexplorer—subagent_type: "sdd:explorer"(fallbackExplore/ inline per../_shared/agent-roster.md) — to localize the code path. Three outcomes (decision table →./references/triage.md):- (a) Regression — an AC describes the expected behaviour and the code violates it. The spec is right; only the code changes.
- (b) Spec-bug — the AC exists but a reasonable implementer could read it and produce the observed behaviour. The wording is the root cause — the AC gets patched (with the user, step 5).
- (c) Gap — no AC covers the behaviour. A new AC is added to §5, marked
<!-- added-by-fix: <date> -->. - No-spec mode — no
docs/features/(or no spec plausibly owns the symptom): skip the spec patch, say so in the record, recommendsurvey.
- RED — pin the bug with a failing test. Write the minimal test that reproduces the bug at the level the behaviour implies (unit for a rule, integration for a dependency behaviour, e2e for a flow). Run it and classify the first run per
../implement/references/tdd-loop.md— it must be a GOOD red (fails on the assertion that encodes the expected behaviour); quote the failing line. A bug that can't be pinned by a test → STOP and say so — an unverifiable fix is a guess. - GREEN + GATE — minimal fix. Make the RED test pass with the smallest change; no drive-by refactors (anything the fix exposes goes to the record's follow-ups). Then the same per-task gate
implementruns: unit + lint + vet (+ integration when available), via the detected commands (detection cascade →../implement/references/command-detection.md). Red gate → fix it, never commit around it. - Spec patch + fix record. Apply the step-2 branch — (a) nothing to patch, re-verify the AC; (b) patch the AC wording; (c) add the new AC with the marker. Any spec change is confirmed with the user in one
AskUserQuestion(before/after wording shown). Then writedocs/features/<slug>/_fixes/<date>-<short-slug>.mdfrom./templates/fix-record.md: symptom → root cause → the pinning test → the spec patch (or why there is none). - Commit + handoff. With the GATE green (this skill's structural self-check), commit per
../_shared/write-and-commit.md— separate commits: the code fix + its pinning test asfix: <slug> <short summary>with trailersSDD-Fix: <date>-<short-slug>andSDD-AC: <id>(when an AC was traced); the record asfix: <slug> record <date>-<short-slug>; the spec patch (if any) asfix: <slug> spec.md <AC id>. A red gate → no commits. Then emit the stage-handoff block per../_shared/handoff.md(utility variant —/clearoptional): What I did (incl. the commits with short SHAs) + Review (the diff,_fixes/<date>-<short-slug>.md, the spec patch if any) + Run next: resume what you were doing; when the fix touched >5 files or crossed a module boundary, recommend/sdd:review <slug>— a recommendation, not a gate.
Definition of Done
- The bug is reproduced by a test that failed before the fix and passes after — GOOD red proven, failing line quoted.
- The gate is clean: unit + lint + vet (+ integration where available).
- The triage outcome is explicit — regression / spec-bug / gap / no-spec — and the matching spec patch is applied (or its absence explained in the record).
docs/features/<slug>/_fixes/<date>-<short-slug>.mdexists: symptom, root cause, the test, the spec patch, follow-ups.- The commit carries the
SDD-Fix:trailer (+SDD-AC:when traced); any spec change was user-confirmed. - The RED-pin (failing test first) + the per-task GATE are this skill's structural self-check (
../_shared/self-check.md); its result is reported in the handoff.
Anti-patterns
- Fixing without a pinning test. «It works now» with no RED proof is a guess that re-breaks silently — the exact failure mode this skill exists to stop.
- Patching code when the spec was the bug. If the AC permitted the behaviour, the wording is the root cause; leave it unpatched and the next implementation reintroduces the bug legally.
- Silent spec edits. Every AC patch/addition is confirmed with the user — the spec is a contract, not a scratchpad.
- Drive-by refactoring. The fix commit is minimal; refactors the fix exposed go to the record's follow-ups, not into the same diff.
- Skipping the gate because the change is «one line». One-line fixes break suites just fine.
- Hard-refusing on a repo without specs. A brownfield bug is this skill's front door — degrade to no-spec mode and recommend
survey, never block. - Writing a parallel test when
_fixes/shows the same symptom was fixed before. That's a recurrence — read the old record and strengthen its test instead.
References & template
./references/triage.md— the symptom→spec trace: grep strategy, the regression / spec-bug / gap decision table, theadded-by-fixmarker, no-spec mode, the recurrence check../templates/fix-record.md— the fix-record scaffold (symptom → root cause → pinning test → spec patch → follow-ups).../implement/references/tdd-loop.md— RED classification (GOOD red / BAD red / false-pass) — the semantics step 3 reuses.../implement/references/command-detection.md— how the step-4 gate commands are resolved (settings override → Makefile → package scripts → language manifests).../_shared/ask-style.md·../_shared/agent-roster.md·../_shared/handoff.md.
Example invocation
User: «/sdd:fix — discounts are applied twice when the user clicks pay twice fast» Skill: intake confirms: expected one discount per order, got two on a double-click (all users, since the checkout-discounts release). Trace:
docs/features/checkout-discounts/spec.mdAC-04 says «a discount is applied to an order at most once» → the code violates it → regression.explorerlocalizes the apply-discount handler (no idempotency check). RED: an integration test posting the same apply twice asserts one discount row — fails withgot 2, want 1(GOOD red). GREEN: guard on the existing uniqueness key; gate clean. Spec: nothing to patch (AC-04 was right). Record_fixes/2026-06-12-double-discount.md; commitfix: checkout-discounts double-applied discount+SDD-Fix:/SDD-AC: AC-04trailers. Handoff: 2 files touched → no review push; resume.
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
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.

