AgentSkillsCN

atft-code-quality

在整个ATFT-GAT-FAN代码库中强制执行lint、格式化、类型检查、测试及安全卫生。

SKILL.md
--- frontmatter
name: atft-code-quality
description: Enforce lint, formatting, typing, testing, and security hygiene across the ATFT-GAT-FAN codebase.
proactive: true

ATFT Code Quality Skill

Mission

  • •Maintain production-readiness by catching regressions before merge or deployment.
  • •Automate formatting, static analysis, and targeted test execution.
  • •Surface remediation paths for failing checks with minimal GPU disruption.

Trigger Phrases

  • •“Run quality gate”, “lint/format the repo”, “execute tests”, “type check”, “pre-commit”.
  • •Requests mentioning Ruff, mypy, pytest, security audit, or dependency hygiene.

Quality Pipeline

  1. •Workspace scan
    • •git status --short
    • •tools/project-health-check.sh --section quality
  2. •Formatting & Lint
    • •ruff check src/ --fix
    • •ruff format src/ tests/
  3. •Type Safety
    • •mypy src/gogooku3 scripts/
    • •pyright if TypeScript integration touched.
  4. •Testing Layers
    • •pytest tests/unit -n auto -q
    • •pytest tests/integration -m "not slow"
    • •python test_short_selling.py --strict for risk module.
  5. •Security & Secrets
    • •pip install bandit safety (once) then bandit -qr src/
    • •detect-secrets scan with baseline security/detect-secrets.baseline.
  6. •Pre-commit Sweeps
    • •pre-commit run --all-files
    • •pre-commit run --hook-stage manual conventional-pre-commit.

Specialized Workflows

Fast Feedback (single file change)

  • •ruff check src/gogooku3/<module>.py --fix.
  • •pytest tests/unit/test_<module>.py -k <case>.
  • •mypy src/gogooku3/<module>.py.

GPU-Sensitive Checks

  • •For CUDA kernels or Torch compile edits:
    pytest tests/integration/test_gpu_training.py::test_compile_path --maxfail=1.
  • •Validate memory usage script: python tools/gpu_memory_report.py --dry-run.

Dependency Hygiene

  • •pip-compile requirements.in (if edited).
  • •python tools/dependency_audit.py --fail-on-critical.
  • •Update lockfiles and note changes in docs/ops/dependency_log.md.

Failure Handling

  • •Lint failure → reference Ruff rule from output, fix quickly; prefer ruff --fix-only RULE.
  • •Type errors → add precise type hints, update mypy.ini only with justification in docstring.
  • •Flaky tests → rerun with pytest -k test_name --lf; mark with @pytest.mark.flaky only when documented in tests/README.md.
  • •Security findings → rotate secrets, update .env.example, notify DevSecOps via security/alerts.md.

Codex Collaboration

  • •Launch ./tools/codex.sh "Perform deep static audit of src/gogooku3" for architectural refactors or elusive bug hunts.
  • •Use codex exec --model gpt-5-codex "Review failing pytest logs and propose fixes" when triaging stubborn CI failures.
  • •Reflect Codex recommendations into permanent lint/type rules and note significant changes in quality/last_quality_report.md.

Exit Criteria

  • •All commands exit 0.
  • •Updated artifacts recorded in quality/last_quality_report.md.
  • •Provide summary + next steps in PR description or change log.