Validate Workflows and Handle Errors

Teaches Claude to read validation errors carefully, understand which are critical vs acceptable false positives, design error outputs and retries correctly, and distinguish between validation passing (necessary) and correctness (sufficient), so workflows fail fast with clear signals instead of silently dropping items.

When validation returns errors, this skill activates and guides the fix: read the error type (missing_required, invalid_value, invalid_reference), apply the fix, re-validate (iterative, usually 2-3 cycles). For workflow errors, it teaches per-node error outputs (onError: continueErrorOutput plus wire main[1]), retryOnFail for flaky external calls, error HTTP response mapping (4xx/5xx codes), and Error Trigger workflows for unattended runs. It also identifies false positives (warnings that are safe to ignore) and explains auto-sanitization (operator structures fixed automatically).

Validation profiles (minimal, runtime, ai-friendly, strict) produce different output; runtime is recommended for pre-deployment. Auto-sanitization runs on every workflow save; it fixes operator metadata but not broken connections or paradoxical states. Validation passing does not guarantee correctness; manual review via the Review Checklist catches silent failures like Merge input mismatches and dropped-item Switches.