Birdview skills audit

中文

Run the read-only inventory from the installed Birdview bundle:

node <skill-root>/scripts/birdview.mjs skills audit --project <project-root> --language en --write <project-root>/.birdview/compatibility-audit.json

Set --language to the language of the current user conversation (zh for Chinese, en for English); the CLI cannot infer conversation language by itself. Always pass it when producing the human-readable Markdown report.

The JSON is deliberately two-layered. skills, duplicateNames and issues are inventory facts; assessment is the required human conclusion. Inventory is not the finished audit. In the same turn, read every discovered SKILL.md and compare at least names, trigger descriptions, explicit invocation requirements, automatic invocation policy, disable-model-invocation, write permissions and approval gates. Then write the assessment using the report contract and render both artifacts:

node <skill-root>/scripts/birdview.mjs skills audit --project <project-root> --language en --assessment <project-root>/.birdview/compatibility-assessment.json --write <project-root>/.birdview/compatibility-audit.json --write-markdown <project-root>/.birdview/compatibility-audit.md

The Markdown report is for people: every pair with overlapping trigger scope must state the triggering language, whether the skills are compatible, need ordering, may conflict or remain unresolved, with confidence, rationale, quoted evidence and a recommended execution order. Keep a relationship unresolved only when the bodies or host rules cannot be read; do not leave an empty unreviewed report merely because the inventory succeeded.