Which checking skill do I want?

Four skills check your project, at different layers. They don't overlap — each answers a question the others can't.

Skill	What it checks	Cost	When to run
/ardd-status	Cross-artifact consistency — conflicts, gaps, draft artifacts, constitution violations, orphaned completion flips, in-flight work. LLM judgment.	Cheap	Before planning. Auto-runs as the final step of most state-changing skills.
/ardd-lint	Structural validity — frontmatter enums, required fields, [artifacts: ...] references, cross-file pointers. Deterministic, no LLM judgment.	Free	Anytime — especially after hand-editing anything in .project/.
/ardd-defects	Artifacts vs. the actual codebase — drift between what an artifact says and what the code does.	Expensive (codebase re-survey)	Periodically, or before major planning.
/ardd-audit	The decisions themselves — simplicity, failure modes, robustness, semantics. Challenges intent.	Moderate	When a design deserves pressure-testing, not just checking.

The shape of the split

Two axes separate them:

• What's compared: /ardd-lint and /ardd-status compare project files against each other (structure and meaning respectively); /ardd-defects compares them against the code; /ardd-audit compares them against good judgment.
• Determinism: /ardd-lint is a script and always says the same thing about the same files. The other three exercise LLM judgment and are worth re-running when context changes.

Where each one's findings go

• /ardd-status → STATUS.md — the re-entry point; always names a recommended next step.
• /ardd-lint → terminal output only; it never writes.
• /ardd-defects → DEFECTS.md — each defect is offered as a fix task by the next /ardd-plan run, exactly once (pull a declined one back with /ardd-plan defect:<id>).
• /ardd-audit → audit.md — a working checklist you resolve directly ([x] done, [-] rejected), with a runnable /ardd-refine command on every concrete suggestion.

What none of them do

None of these capture your observations from using the built thing — that's /ardd-feedback. And none vet a new idea — that's /ardd-research's proposal mode.