Labali Video Desensitize

labali-video-desensitize

Treat this skill as a layered system.

Layer Contract

1. SKILL.md is the policy layer.
2. references/architecture.md is the strategy layer.
3. scripts/sanitize-video.sh is the execution layer.

NEVER

• Never skip the two-pass encode — single-pass re-encode does not reliably strip all metadata.
• Never claim full forensic erasure — this method is practical-risk reduction only; state this explicitly in the result.
• Never overwrite or delete the input file — always write to a separate output path.

Pipeline Steps

Execute in fixed order:

Step 1 — Validate prerequisites

• Confirm ffmpeg, ffprobe, and exiftool are available in PATH.
• Fail fast with explicit error if any is missing.
• Confirm input video path is readable and sufficient disk space exists for two-pass re-encode.

Step 2 — FFmpeg two-pass re-encode

• Run scripts/sanitize-video.sh with input and output paths.
• Pass 1: transcode to intermediate mp4 with watermark-resistance transforms:
  • -vf "scale=trunc(iw*0.98/2)*2:trunc(ih*0.98/2)*2,scale=trunc(iw/0.98/2)*2:trunc(ih/0.98/2)*2"
  • -c:v libx264 -crf 28 -c:a aac -b:a 128k
• Pass 2: sanitize from intermediate to output:
  • -map_metadata -1 -map_chapters -1
  • -fflags +bitexact -flags:v +bitexact -flags:a +bitexact
  • -movflags +faststart -write_tmcd 0
• Remove intermediate file after pass 2 completes.

Step 3 — Diagnostics

• Print before/after comparison:
  • file size,
  • duration,
  • dimensions,
  • EXIF line count,
  • SHA256 hash,
  • EXIF diff,
  • ffprobe format-tag diff.

Step 4 — Reviewer: sensitive metadata scan

• Run post-sanitize hidden metadata scan on output video via scripts/check-sensitive-video-info.sh.
• Check EXIF/container metadata for sensitive key/value patterns.
• Print Sensitive info review: PASS|REVIEW_REQUIRED.

Step 5 — Return result

• Load references/risk-notes.md before writing the result caveats to ensure the framing is accurate.
• State clearly: this method is practical-risk reduction, not 100% forensic erasure.
• State clearly: default stronger transforms improve disruption odds for platform watermarking but cannot guarantee full removal.
• In normal mode: report verdict and summary.
• In --strict mode: exit non-zero when suspicious metadata candidates are detected.

Success Criteria

A run is successful only when all conditions hold:

1. Input video is readable.
2. Output video is generated at target path.
3. Before/after diagnostics are printed.
4. EXIF and ffprobe diff sections are printed (empty diff is acceptable).
5. Post-sanitize sensitive metadata scan section is printed.
6. In strict mode, suspicious hidden metadata produces non-zero exit.
7. Intermediate file is removed after completion.

Runtime Inputs

Use skill.yaml as input schema source of truth.

Operational Mode

• Deterministic local shell execution only.
• No browser, no network, no API.
• Keep output reproducible and script-driven.
• State clearly that this method is practical-risk reduction, not 100% forensic erasure.
• State clearly that default stronger transforms improve disruption odds for platform watermarking, but cannot guarantee full removal.

Resources

• Architecture: references/architecture.md
• Workflow plan: references/plan.md
• Risk notes: references/risk-notes.md
• Deterministic script: scripts/sanitize-video.sh
• Post-check script: scripts/check-sensitive-video-info.sh
• Optional CLI wrapper: scripts/run.ts
• Regression checks: tests/test_regression.sh