ground-readmelisted

# Ground README — Audit or generate honest project listings You are either auditing an existing README/marketplace listing for overpromise and unbacked claims, or generating a new one from actual code with grounding controls. This is the **README-shaped version of the rad-explain principle**: every substantive claim in a project's outward-facing copy should be checkable against source, and nothing should use marketing/sensational language. The principle is the same one the rad-claude-skills marketplace applied to its own listings post-audit. ## Foundational rules 1. **Grounded.** Every claim about what the project does must trace to code, docs, or test evidence. The `check-grounding.py` validator runs against the output. 2. **Not overpromising.** No "the only" / "world-class" / "revolutionary" / "production-grade" without evidence. The `check-overpromise.py` validator runs against the output. These are non-negotiable. If they fail, the skill surfaces findings and offers revision before finalizing. ## Modes ### `audit` (most common) Review an existing README, marketplace listing, plugin.json description, or similar outward-facing copy. Produce a findings report. Workflow: 1. Read the target file (default: `README.md` at repo root) 2. Run `check-overpromise.py` against it 3. Run `check-grounding.py` against the repo 4. Read the actual code/docs to verify capability claims 5. Surface findings with line numbers + suggested rewrites 6. Optionally write a revised version