Summary
The README is stale — it reflects the initial release but not the verification criteria system, domain criteria, eval harness, or the docs/tests infrastructure added since. Install instructions and compatible model/platform claims also need verification against current behavior.
What needs updating
1. Document new features
- Verification criteria system — domain-specific audit criteria (e.g., Implementation Plans) that harden loads and applies
- Domain criteria files — extensible criteria in
tests/fixtures/ and the pattern for adding new domains
- Eval harness —
tests/run-eval.md and the fixture-based evaluation system for testing harden itself
- Anti-pattern detection — harden now catches known failure modes (e.g., static-only audits)
2. Verify install instructions
- Confirm claude.ai ZIP upload path still works (Settings → Capabilities → Add Skills)
- Confirm Claude Code install path (
.claude/skills/) is accurate
- Verify Codex CLI / Gemini CLI paths are correct or update
- Test that the skill activates after a fresh install via at least one method
3. Verify applicable models / platforms
- Confirm
compatibility field in SKILL.md frontmatter is accurate (claude.ai, Claude Code, Codex CLI, Gemini CLI)
- Note any platform-specific limitations (e.g., web search availability)
- Verify graceful degradation claim (works without web search)
4. Reflect repo structure
- README should acknowledge
docs/, tests/, and CONTRIBUTING.md exist — users landing on the repo should know where to look
Acceptance criteria
Likely affected files
README.md
SKILL.md (compatibility field, if corrections needed)
Summary
The README is stale — it reflects the initial release but not the verification criteria system, domain criteria, eval harness, or the docs/tests infrastructure added since. Install instructions and compatible model/platform claims also need verification against current behavior.
What needs updating
1. Document new features
tests/fixtures/and the pattern for adding new domainstests/run-eval.mdand the fixture-based evaluation system for testing harden itself2. Verify install instructions
.claude/skills/) is accurate3. Verify applicable models / platforms
compatibilityfield in SKILL.md frontmatter is accurate (claude.ai, Claude Code, Codex CLI, Gemini CLI)4. Reflect repo structure
docs/,tests/, andCONTRIBUTING.mdexist — users landing on the repo should know where to lookAcceptance criteria
Likely affected files
README.mdSKILL.md(compatibility field, if corrections needed)