build-debug-tool
/build-debug-tool
Zamienia powtarzane przeszukiwanie logów i zrzutów w stałe narzędzie diagnostyczne, gdy tylko zauważysz problem.
- Autor
- deepklarity ↗
- Repozytorium
- harness-kit ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 10.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zamienia powtarzane przeszukiwanie logów i zrzutów w stałe narzędzie diagnostyczne, gdy tylko zauważysz problem.
Wyzwalacze z opisu
Oryginalny opis (EN)
Turn a repeated hunt through logs, JSON dumps or database rows into a permanent diagnostic tool, at the moment you notice the pain. Triggers on: 'build a debug tool', 'make this dump readable', 'I keep grepping for this', 'I keep typing this python -c', or /build-debug-tool. Test commands and test data belong to hk-testing-shortcuts.
Jak działa
- The tells
- Where the tool goes
- The bar
- Example
Podgląd SKILL.md
---
name: build-debug-tool
description: "Turn a repeated hunt through logs, JSON dumps or database rows into a permanent diagnostic tool, at the moment you notice the pain. Triggers on: 'build a debug tool', 'make this dump readable', 'I keep grepping for this', 'I keep typing this python -c', or /build-debug-tool. Test commands and test data belong to hk-testing-shortcuts."
---
# build-debug-tool
Build the tool the moment you are reading more than the answer needs, retyping the same extraction, or chaining grep and python over JSON to answer a question you have asked before. A permanent, named tool; raw logs and ad-hoc scripts are the fallback for when the tool itself looks wrong, never the routine. Before this workspace's diagnostic scripts existed, one session spent an hour grepping escaped log lines and retyping the same ten-line python three times; the scripts then answered every question in seconds.
## The tells
Any one of these means stop and build:
- You typed nearly the same extraction twice (a grep chain, a `python -c` over JSON, a jq pipe).
- You are reading a wall of output to find one fact (a giant debug dump, an escaped multi-line error, a 300-line JSON).
- A check failed and the reason lives somewhere you cannot read with one command.
- You wrote a "temporary" script in the scratchpad that you would want again tomorrow.
## Where the tool goes
Look first at what exists: the workspace's diagnostic-scripts folder, its run/log command's help, and its repo-wide scripts folder (CLAUDE.md names each). Extend one of those when it already answers part of the question; a new mode or `--sections` entry on an existing script beats a new script.
1. **A service's own data** (its entities, records, jobs): a script in that service's diagnostic-scripts folder (CLAUDE.md), following the shared-helper pattern its neighbours use so it gets `--brief`, `--json`, `--full` and `--sections`. It reads the data directly (through the app's own data layer, or a direct query) rather than over the network when the network path needs auth the script can bypass. Add a row to the "when to reach for which script" table in the root `CLAUDE.md`.
2. **Orchestration runs and logs**: a flag on the workspace's run/log command or CLI. Log extraction is the right answer only when the fact cannot be made to land somewhere a script can read it (an error inside a rolled-back transaction never reaches the run's own record).
3. **Repo-wide checks and comparisons**: a script in the repo-wide scripts folder, the shape of the verify gate: one command, a one-line verdict, evidence attached on failure.
4. **Test commands, test IDs, how-do-I-test knowledge**: `hk-testing-shortcuts` owns it; call the Skill tool with `hk-testing-shortcuts` and stop here.
## The bar
- One command, one question answered, one screen of output.
- Works from the ID forms a person actually has (a task id or its board key, a spec id or its slug).
- Prints the denominator with the fact ("comments: (none)" beats empty output).
- A time budget where hanging is possible; on failure, print the evidence, not just the failure.
- Stays in the same change as the work that needed it, so it is committed together when the user commits. An uncommitted tool is a snippet.
Done when the tool answers the question that caused the pain in one command on the real case, and the chat reply gives that command line.
## Example
- Existing tools: list your testing-tools folder, your CLI's `--help`, and `scripts/`.
- A data script (records, boards, reports) lives in the diagnostic-scripts folder and reads through the app's own data layer; CLAUDE.md rules out curl and raw API calls.
- A CLI log tool is a flag on the existing log command.
- A repo-wide script has the shape of `scripts/verify.sh`.
Wersje
W bazie nie ma innych skilli o nazwie build-debug-tool.
Migawka bazy z 13.11.2026 · wersja paczki 42ecdaaaa263 (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
- 98ci-crash-dump
Pobiera i analizuje zrzuty awarii z testów CI w repozytoriach dotnet, gdy test się wywrócił, a nie tylko nie przeszedł, na przykład w pull requeście. - 100deepgram-debug-bundle
Zbiera dowody diagnostyczne Deepgram dla wsparcia technicznego: przygotowanie zgłoszeń, badanie problemów i zbieranie logów oraz informacji diagnostycznych. - 100deepgram-debug-bundle
Zbiera dowody diagnostyczne Deepgram dla wsparcia technicznego: przygotowanie zgłoszeń, badanie problemów i zbieranie logów oraz informacji diagnostycznych. - 100groq-debug-bundle
Zbiera dowody diagnostyczne Groq do zgłoszeń wsparcia i rozwiązywania problemów, gdy trwałe kłopoty wymagają logów i informacji diagnostycznych. - 100groq-debug-bundle
Zbiera dowody diagnostyczne Groq do zgłoszeń wsparcia i rozwiązywania problemów, gdy trwałe kłopoty wymagają logów i informacji diagnostycznych. - 95tarn-api-testing
Pisze i uruchamia testy API w Tarn, w plikach .tarn.yaml, a także generuje testy z OpenAPI lub poleceń curl.