SuperAPI AI Directory
Partner中文AI newsFavorites
Skillstech-debt

tech-debt

Testing & quality

View repository

Updated 10-17 21:01

About this skill

Publisher source snapshot: 2026-10-17 UTC, fixed Git revision. The complete original SKILL document follows. Companion reference files were not fetched; commands were not executed.

Technical Debt

Operations

  • research [scope]: inspect the requested scope, defaulting to turbo/, and report verified findings. This operation is read-only.
  • issue [report]: create an English GitHub issue from current, verified findings. If no report exists, research first. Create or comment on an issue only when the caller requested that operation.

The tech-debt-research and tech-debt-issue commands remain aliases for these operations. A report is evidence at its recorded revision, not a standing instruction to change code.

Research

  1. Record the repository, HEAD, scope, and date. Read code quality and use the index for the affected surfaces.
  2. Use rg to find candidates and inspect matching files and consumers. Exclude dependencies, generated/vendor content, and historical migrations from generic cleanup recommendations. Preserve permanent migration records.
  3. Check type/lint suppressions, dynamic imports, environment configuration, error handling, fallback ownership, and unused dependencies against actual contracts. Use testing guidance for tests and ccstate guidance for signal/React code.
  4. Validate each candidate. File length, a relative import, a catch, or an ESLint off / Oxlint allow setting is not proof of a defect. Read override scope, replacement enforcement, generated-code boundaries, and documented exceptions. In particular, do not replace every floating promise with detach() or assume every parentless resetSignal() leaks.
  5. Report the concrete consequence, file/line, applicable rule, evidence, proposed remedy, and uncertainty. Use severity based on demonstrated impact. Distinguish static inspection from executable or production verification.

Keep the report proportional to the findings: scope and revision, confirmed issues, dismissed candidates when useful, and next actions. Save detailed evidence when it is too long for the response. Do not invent schedules or effort estimates from match counts.

Issue

Recheck the report against current source and existing issues before posting. Use an English title describing the concrete problem; include affected paths, evidence, impact, and actionable acceptance criteria in the body. Use only existing relevant labels and include source links pinned to the inspected SHA.

Write the body to a temporary file and pass gh issue create --repo okou-ai/okou --body-file <file> the exact Markdown. Keep the report focused enough to fit one issue; add detailed comments only when needed for the requested tracking task. Verify the created issue and return its URL. If posting fails, retain the report and state which action failed.