agentleFS
mocchaleramocchalera/video-os-v2/.agents/skills/finish-interview/SKILL.md

finish-interview

Skill4 starsChanged 2 months agoMITView on GitHub

Apply repeatable dialogue MA, Studio-based or automatic portrait reframing, overlay-safe rendering, and loudness/sync QA to an existing interview or talking-head rough cut. Use when the user asks to make speech easier to hear, perform MA, enlarge or rebalance a person in frame, crop or punch in an interview, automatically propose framing, or prepare a finishing preview before final approval.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use canonical timeline metadata and the shared renderer.

    line 8
  2. Very specific:

    Read 05_timeline/timeline.json, 06_review/review_patch.json, the latest preview, and any downstream card/caption artifacts.

    line 14
  3. Distinctive:

    create a representative frame contact sheet.

    line 17
  4. Very specific:

    when choosing framing automatically, run .build/debug/videoos-studio-cli interview-reframe --source=<video> --in-us=<n> --out-us=<n>.

    line 22
  5. Very specific:

    treat its face/eye/yaw/hand result as a proposal, then inspect representative frames.

    line 24
  6. Distinctive:

    Keep editorial timing unchanged unless the user separately requests a cut.

    line 25
  7. Distinctive:

    Write review patch operations:

    line 32
  8. Distinctive:

    Verify the final preview by full decode, loudness measurement, visual frame sampling, audio/video duration comparison, and dialogue …

    line 42
  9. Distinctive:

    No cropped head, hands, or interviewer look room in representative samples.

    line 97
  10. Distinctive:

    Keep preview output distinct from final package output until caption and review gates are approved.

    line 114

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Finish Interview

Use canonical timeline metadata and the shared renderer. Do not handwrite a one-off ffmpeg graph unless the project's downstream overlay format cannot be represented by the timeline.

Workflow

  1. Read 05_timeline/timeline.json, 06_review/review_patch.json, the latest preview, and any downstream card/caption artifacts.
  2. Measure before changing:
    • create a representative frame contact sheet;
    • measure integrated LUFS, true peak, and LRA;
    • probe video/audio start times and durations;
    • record avg_frame_rate, r_frame_rate, and time_base; treat a phone source whose cadence cannot be proven CFR as VFR for sync review.
    • when choosing framing automatically, run .build/debug/videoos-studio-cli interview-reframe --source=<video> --in-us=<n> --out-us=<n>;
    • treat its face/eye/yaw/hand result as a proposal, then inspect representative frames.
  3. Keep editorial timing unchanged unless the user separately requests a cut. For a titled J-cut opening, stage information instead of presenting exterior, title, and dialogue at the same instant: establish the exterior, let the title read, then fade dialogue in before the title exits. Enter the interview on a moving source frame, never a frozen bridge frame. Preserve the complete final utterance and a short moving-picture tail; do not extend the ending with a freeze frame.
  4. Write review patch operations:
    • one change_visual_transform per affected video clip;
    • one timeline-level change_audio_finish operation. Studio can author both operations from the selected clip's インタビュー仕上げ inspector. Adjusted framing appears immediately in the Viewer; press 画角を保留 and MAを保留 before Apply & Preview.
  5. Apply the patch with scripts/compile-timeline.ts --patch and require schema validation to pass.
  6. Render the clean assembly first. Burn captions and insert question cards afterward so zoom/crop never scales authored graphics.
  7. Verify the final preview by full decode, loudness measurement, visual frame sampling, audio/video duration comparison, and dialogue placement QA. The duration comparison alone is insufficient: package QA must show dialogue_timeline_alignment_valid as well as av_drift_valid.

Patch pattern

{
  "timeline_version": "1",
  "operations": [
    {
      "op": "change_visual_transform",
      "target_clip_id": "CLP_0001",
      "visual_transform": {
        "zoom": 1.15,
        "position": { "x": -144, "y": -39 }
      },
      "reason": "Increase portrait presence while preserving look room"
    },
    {
      "op": "change_audio_finish",
      "audio_finish": {
        "preset": "dialogue-clean",
        "loudness_target_lufs": -16,
        "true_peak_target_dbtp": -1.5
      },
      "reason": "Improve speech intelligibility and delivery loudness"
    }
  ]
}

position is measured in output pixels. Negative x shifts the image left; negative y shifts it upward. Start around zoom: 1.10–1.18 for a restrained talking-head punch-in and inspect hand gestures across the whole cut.

For zoom > 1, position pans inside the available zoom overscan and clamps at the source boundary. Do not recreate the old post-crop pad/translate graph; it exposes black edges. The automatic planner normally caps an interview punch-in at 1.18, reduces it when hands are detected, and allows stronger zoom only when the detected face is genuinely small.

Audio presets

  • dialogue-clean: 70 Hz high-pass, gentle noise reduction, mud/presence EQ, 3:1 compression, and measured two-pass loudness normalization.
  • loudness-only: measured two-pass normalization without cleanup or EQ.
  • none: disables the declared finish preset.

Defaults target -16 LUFS, LRA 7, and -1.5 dBTP. The dialogue preset keeps 0.3 dB of codec headroom before AAC encoding.

QA contract

  • No cropped head, hands, or interviewer look room in representative samples.
  • Captions and question cards retain their authored size and position.
  • Integrated loudness is within 0.5 LU of target.
  • Encoded true peak does not exceed the declared target by more than 0.2 dB.
  • Audio/video start at zero and end within one frame.
  • Dialogue-only signal stays inside the timeline windows that own dialogue. adelay=...,atrim=start=0 on one input branch is forbidden because the trim removes the inserted lead-in; source atrim must precede adelay, and only the final mixed output may be duration-trimmed.
  • For VFR phone footage, first choose the intended video frame, then measure the video and audio source offsets independently. If frame quantization leaves a sub-frame residual, keep the chosen picture frame and offset audio by the measured residual. Do not chase sync by repeatedly moving the video trim across frame boundaries.
  • A proxy or alternate encode made to diagnose a player is diagnostic only. It must not become the canonical creative output or constrain title/J-cut timing.
  • Full ffmpeg decode exits successfully.
  • Keep preview output distinct from final package output until caption and review gates are approved.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.