@promptci/cli

Instruction health for AI coding workflows. Scans AI coding instruction files (CLAUDE.md, AGENTS.md, .cursorrules, Copilot instructions, and more) and produces actionable health reports: duplicates, conflicting directives, stale guidance, context bloat, broken references, and vague instructions.

The scanner is deterministic and rule-based — no LLM calls, no network access, identical output for identical input.

Usage

npx @promptci/cli scan

Writes .promptci/latest.md (human-readable) and .promptci/report.json (machine-readable), and prints a summary with a health score and top fixes.

Common commands:

npx @promptci/cli scan --path /path/to/repo   # scan a specific repo
npx @promptci/cli init                        # create .promptci/config.json
npx @promptci/cli fix                         # apply deterministic fix recipes
npx @promptci/cli review-diff --base origin/main   # CI: fail on instruction regressions

Full documentation: https://github.com/SeriousGeese/PromptCI

License

Apache-2.0. The scanner engine lives in @promptci/core.