Fix bugs and broken behavior when there is enough evidence to act on a repair path. Use for errors, crashes, incorrect results, API failures (500, 404, 403), CORS problems, database exceptions, broken rendering, duplicated or wrong data, off-by-one mistakes, timezone/date bugs, broken forms, config-caused runtime failures, and regressions. Trigger when the user wants the bug repaired and the conversation already contains a clear failing area, a reproducible failing test, a concrete error path, or a prior diagnosis to implement. Do NOT use for new features, pure explanation, architecture discussion, broad research, or bug reports where the main need is figuring out why the behavior happens — use diagnose for that.
npx skills add https://github.com/avibebuilder/claude-prime --skill fix
Think harder.
Remove the cause with the change that genuinely restores the intended behavior. Making the symptom disappear without explaining the evidence is not a fix.
Check conversation context and skip completed steps.
Read the symptom, expected behavior, errors, logs, failing tests, and any prior diagnosis. Separate confirmed facts from guesses. Then route:
| Situation | Action |
|-----------|--------|
| Clear root cause or one strongly evidenced failing area | Stay in /fix |
| One narrow check would remove the last uncertainty | Do that check inside /fix, then commit to a lane |
| Multiple plausible causes, unclear failing area, or needs runtime instrumentation | Switch to /diagnose first |
| Bug is understood but multiple defensible fixes with real tradeoffs | Switch to /discuss |
If you're about to add a speculative guard or workaround because the cause is still fuzzy, you're in the wrong lane. If evidence is insufficient, switch to /diagnose instead of guessing.
GATE: If a plan was requested or produced, wait for user approval before implementation.
After 3 substantive fix attempts that haven't resolved the bug, stop thrashing. Write a handoff note covering: bug context, confirmed evidence, files checked, each failed approach and why it failed, open questions, and most likely next diagnostic branch. Start a fresh Claude session with the handoff note (or give it to the user to paste). Repeated failures signal contaminated context or narrowed reasoning — a clean window gets fresh judgment. Let stop and enjoy the world, you just did the best thing bro!
undefined, null, wrong type), trace back to where that data was produced or passed. Fix the producer or caller, not the victim. Example: applyDiscount(cart, coupon) crashes because coupon is undefined → fix the lookup or call site that passed bad data, not applyDiscountA repair is only done when the evidence matches the report. Prove three things: (1) the original failure is gone, (2) the repaired path was actually exercised, and (3) nearby behavior did not regress.
Hand off to a tester — an isolated teammate that verifies the repair independently. See .claude/skills/test/teammate.md for how to spawn one.
Add or update a durable test in /fix when covering the bug clearly belongs in the codebase. Otherwise the tester owns verification.
Remove temporary debugging artifacts once verification passes: throwaway scripts, temp logs, or ad hoc instrumentation. Keep durable tests and intentional logging.
GATE: Do not call the bug fixed until the evidence directly addresses the reported failure.
<issue>$ARGUMENTS</issue>
LLM-driven hypothesis generation/testing on tabular data. Three methods: HypoGeniC (data-driven), HypoRefine (literature+data), Union. Iterative refinement, Redis caching, multi-hypothesis inference. Manual: hypothesis-generation; ideation: scientific-brainstorming.
This skill should be used when the user asks to \"automate SQL injection testing,\" \"enumerate database structure,\" \"extract database credentials using sqlmap,\" \"dump tables and columns...
Optimizes application performance across frontend, backend, queries, and databases. Use when performance requirements exist, when you suspect performance regressions, when Core Web Vitals or load times need improvement, when N+1 query patterns need fixing, or when profiling reveals bottlenecks.
Bisect a ClickHouse regression using pre-built master binaries from CI. Use when the user wants to find the commit that introduced a bug.
QA an analysis before sharing -- methodology, accuracy, and bias checks. Use when reviewing an analysis before a stakeholder presentation, spot-checking calculations and aggregation logic, verifying a SQL query's results look right, or assessing whether conclusions are actually supported by the data.
End-to-end smoke test for the public Errors HTTP API (error groups). Seeds failed runs into ClickHouse so the error materialized views populate, then drives the real endpoints against the running webapp — list (with filters + pagination), retrieve, resolve/ignore/unresolve, the `filter[error]` runs filter, user attribution via the `trigger.dev mint-token` -> JWT exchange, and the 401/403/404 negatives. Use for "smoke test the errors API", "test the errors API e2e", "prove the errors endpoints work", or to re-verify after changes.
Strix SQL 注入测试手册,覆盖 union、blind、error-based 与 ORM 绕过技巧;触发名:strix-sql-injection
> Pull and interpret production experiment query-performance data from the staff-only slowest experiment queries, precompute read/build health, and preaggregation cache footprint. and response field semantics (exception codes, exposure paths, precompute skip reasons, job states). Use when investigating slow or failing experiment queries, precompute regressions, 307/159/241 errors, preaggregation table growth, or when asked how experiment query performance or the precompute rollout is doing in production.
Take avibebuilder/fix from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.