audit-bugs
/audit-bugs
Szuka prawdziwych błędów w kodzie, np. logicznych, odwołań do null i wycieków zasobów, oraz debuguje konkretne objawy.
- Autor
- paulnewmandev ↗
- Repozytorium
- pn-ultra-audit ↗
- Licencja
- GPL-3.0
- Ostatnia zmiana repozytorium
- 13.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.
GPL-3.0, otwarta.
Co robi i kiedy
Szuka prawdziwych błędów w kodzie, np. logicznych, odwołań do null i wycieków zasobów, oraz debuguje konkretne objawy.
Wyzwalacze z opisu
Oryginalny opis (EN)
Bug locator. Finds real defects (logic errors, null/undefined dereferences, race conditions, resource leaks, bad error handling, off-by-one, timezone/money/rounding errors, async misuse, broken edge cases, dead branches) by tracing code paths, and can also debug a specific symptom (stack trace, failing test, wrong output) down to root cause. Use for "find bugs", "why does X fail", "locate the bug", "revisar errores", or as the bug pass of /audit.
Jak działa
- Mode A — Sweep (no specific symptom)
- Mode B — Symptom debug (error/trace/test/wrong output given)
- Sweep method
- Evidence bar
- Severity quick guide
- Do not report
- References
Podgląd SKILL.md
--- name: audit-bugs description: Bug locator. Finds real defects (logic errors, null/undefined dereferences, race conditions, resource leaks, bad error handling, off-by-one, timezone/money/rounding errors, async misuse, broken edge cases, dead branches) by tracing code paths, and can also debug a specific symptom (stack trace, failing test, wrong output) down to root cause. Use for "find bugs", "why does X fail", "locate the bug", "revisar errores", or as the bug pass of /audit. --- # audit-bugs Two modes. Pick by input. ## Mode A — Sweep (no specific symptom) Hunt unknown defects. Output candidates in `../audit/references/finding-schema.md` with `type: "bug"`. ## Mode B — Symptom debug (error/trace/test/wrong output given) Systematic, hypothesis-driven. Do not guess-fix. 1. **Restate** symptom precisely: expected vs actual, where observed, since when (`git log -S`, `git bisect` plan if regression). 2. **Reproduce** locally if possible (existing test, minimal script). If it can't be reproduced, say so and list what's needed. 3. **Trace backwards** from the failing line along data flow; list ≥ 2 hypotheses, state what evidence would confirm each, then test them cheapest-first (read code, add targeted assertion/log in a scratch copy, run single test). 4. **Root cause statement**: one sentence, "X happens because Y at file:line, when Z". 5. **Fix** at the root, not the symptom; write/describe the regression test that fails before and passes after. 6. **Blast radius**: grep for the same pattern elsewhere (siblings of the bug) and report them as additional findings. ## Sweep method 1. Take ledger units for `audit-bugs` (hot modules first: money, auth, state machines, parsers, schedulers, sync/import/export, concurrency, date/time). 2. If installed, run available static checkers and **read their output critically** (they produce noise): `tsc --noEmit`, `eslint`, `phpstan`/`psalm`, `mypy`/`ruff`, `go vet`/`golangci-lint`, `cargo clippy`, `semgrep --config auto`. Keep only issues you can trace to a concrete failing input. 3. Read the code of each unit. For each function ask the **five questions**: - What inputs reach this (types, null, empty, huge, negative, unicode, concurrent)? - What does it assume that the caller does not guarantee? - What happens on the error/timeout/partial-failure path? - What shared state does it touch, and who else touches it? - What does the test suite *not* cover here? 4. Use `references/bug-patterns.md` as a prompt list, not a checklist to tick. A pattern match is a *candidate* until you construct a concrete failing input. 5. **Construct the failing case**: give the exact input/state/sequence that triggers it and the wrong result. If you can't, it's `needs_validation` or dropped. 6. Grep for siblings of each confirmed bug. ## Evidence bar A bug finding needs: location, quoted code, the **triggering input or interleaving**, actual vs expected behavior, and impact (who sees wrong data/crash/loss). No "this looks fragile". ## Severity quick guide - critical/high: data loss/corruption, wrong money/quantities, crash of core path, auth-dependent logic wrong, infinite loop/hang on normal input. - medium: wrong result in less common but realistic cases; resource leak that exhausts under load; swallowed errors hiding failures in important flows. - low: wrong result only in rare edge; cosmetic mis-behavior. ## Do not report Style, naming, missing docs (→ audit-code); unproven theoretical races; intended behavior you merely dislike; framework-guaranteed invariants; test-only code unless it hides a real failure. ## References - `references/bug-patterns.md`
Pliki w paczce (2)
- SKILL.md 4 KB
- references/bug-patterns.md 6 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · paulnewmandev/pn-ultra-audit | 100 | GPL-3.0 |
| TalonT-Org/AutoSkillit | 100 | MIT |
| TalonT-Org/AutoSkillit | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki cc0ab6c0f5d1 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100bug-detector
Wykrywa błędy, błędy logiczne i problemy z przypadkami brzegowymi w kodzie, w tym błędy wykonania, obsługę null i undefined oraz błędy typów, i zwraca raporty ze ścieżkami, numerami linii i poprawkami. - 100code-review
Sprawdza kod lub diff pod kątem błędów logicznych, problemów z bezpieczeństwem, obsługi błędów, współbieżności i czytelności oraz wskazuje możliwe uproszczenia. - 100debug-detective
Szuka przyczyny błędu, awarii lub nieudanego testu na podstawie stack trace'u i naprawia ją u źródła. - 100debug-root-cause
Szuka i naprawia systematycznie przyczyny błędów, awarii i nieudanych testów w JavaScript, TypeScript i Pythonie, zamiast zgadywać poprawkę. - 95static-bug-detector
Analizuje kod statycznie pod kątem błędów funkcjonalnych, takich jak null dereference, błędne warunki, nieosiągalny kod i wycieki zasobów, z opisem, poziomem ważności i oceną pewności. - 100bug-hunter
Bada błędy i awarie na podstawie logów i stack trace, proponuje minimalne poprawki i testy regresyjne.