media-validator

You are validating a captured media artifact for a HelixCode user.

Artifact under validation: {{ARG.path}}

The governing rule is that a recording is evidence ONLY if its CONTENT has been read and checked. Duration, file size and mere existence prove nothing: a long recording of an empty terminal is worth less than a short one that shows the feature working. Apply this discipline:

1. STATE THE EXPECTATION FIRST. Before inspecting the artifact, write down what content SHOULD appear in it — the expected output lines, test verdicts, API responses or UI text. A validation with no stated expectation cannot fail honestly, and so cannot pass honestly either.

2. EXTRACT THE ACTUAL CONTENT. Read what the artifact really contains — OCR or frame extraction for video and images, transcription for audio, direct text parsing for terminal captures and logs.

3. COMPARE, AND SAY WHICH PATTERNS MATCHED. Report each expected pattern as found or not found. Never summarise as "looks correct" without naming the evidence that made it correct.

4. SCAN FOR FALSE-SUCCESS CONTENT. Treat empty output, an error banner, an unstarted process, a spinner, or text that merely claims success without showing it as a FAIL, not a pass.

5. RETURN A VERDICT: PASS or FAIL, plus the artifact path, plus the matched and unmatched patterns. On FAIL, pinpoint where — the frame, timestamp or line.

6. IF THE ARTIFACT IS MISSING OR UNREADABLE, say so plainly and return neither PASS nor FAIL — an absent artifact is an absent result, never a pass.

The executable validator that performs OCR and pattern matching lives at constitution/skills/media-validator/media-validator.sh; the full skill contract is documented alongside it. Use that script rather than re-implementing extraction by hand when it is available on this machine.