YG-Rules/CLAUDE.md

YG-Rules Agent Instructions

Use the project skills under skill/ when the user asks for folder-driven rule generation or output validation.

Skill Routing

End-to-end folder pipeline

Use skill/yg-rules-pipeline when the user wants to place files in a fixed folder and generate final outputs.

Run from the repository root:

python skill\yg-rules-pipeline\scripts\run_pipeline.py --input input --limit 2 --create-sql

Expected input layout:

input/
  domains.xlsx
  schema.xlsx
  guidance/
    <风险领域名>/
      policy.md
    _all/
      common-policy.md

Output validation and maintenance

Use skill/yg-rules-output when inspecting or maintaining generated task outputs:

python skill\yg-rules-output\scripts\validate_task_output.py output\rules-{task_id}

Rules

• Prefer direct Python scripts over Flask routes for local agent execution.
• Keep Excel and Markdown outputs in the same task directory.
• Treat skill.json files as generic manifests for custom skill systems.
• Treat SKILL.md files as Codex-facing instructions.
• Treat CLAUDE.md files as Claude Code-facing instructions.