igapyon-ffmpeg-helper

Files (12)

• SKILL.md
• openai.yaml · agents/
• youtube-output-policy.md · references/decisions/
• audio-concat.md · references/process/
• audio-trim.md · references/process/
• delivery-variants.md · references/process/
• h4essential-input-discovery.md · references/process/
• peak-gain-normalize.md · references/process/
• still-image-youtube-video.md · references/process/
• workspace-and-command-log.md · references/process/
• youtube-manual-upload.md · references/process/
• h4essential-orchestra-youtube.md · references/workflows/

SKILL.md (first lines)

# igapyon-ffmpeg-helper

This skill guides igapyon's personal FFmpeg workflows. Use it as a
conversational command generator and runbook navigator.

Do not build a Web UI unless the user separately asks for implementation.

## Activation

Use this skill only on a hard trigger. Valid triggers include:

- `igapyon-ffmpeg-helper`
- `igapyon FFmpeg helper`
- `ffmpeg helper`
- `FFmpeg helper workflow`
- `FFmpeg helper runbook`
- `ffmpeg runbook igapyon`
- `igapyon ffmpeg runbook`
- `igapyon's FFmpeg helper workflow`
- an explicit request to use or apply this Agent Skill

Do not activate this skill for:

- ordinary FFmpeg questions
- ordinary H4essential questions
- ordinary YouTube/video/audio conversion questions
- mentions of `loudnorm`, gain adjustment, trimming, concatenation, or still
  image video creation
- vague descriptions of a similar workflow

If the user only asks whether such a skill exists, mention this skill as an
available option, but do not apply it until the user asks to use it.

## Workflow Selection

The first-cut workflow is:

- [references/workflows/h4essential-orchestra-youtube.md](references/workflows/h4essential-orchestra-youtube.md)

When more workflows are added, choose the closest workflow from
`references/workflows/`. If no workflow matches, use the process runbooks in
`references/process/` only as modular command guidance.

For the first-cut workflow, confirm the route once at the start:

```text
H4essentialのオーケストラ録音から、切り出し、loudnorm JSON測定、元のサンプルレート・32-bit floatを保った-0.5 dBTPピーク基準の単純ゲイン仕上げ、必要なら結合、静止画付きYouTube用動画作成まで進めますか？
```

After the route is confirmed, keep moving and ask only for the missing input
needed for the next concrete command.

## Process Runbooks

Use these process files as reusable command building blocks:

- [references/process/h4essential-input-discovery.md](references/process/h4essential-input-discovery.md)
- [references/process/workspace-and-command-log.md](references/process/workspace-and-command-log.md)
- [references/process/audio-trim.md](references/process/audio-trim.md)
- [references/process/peak-gain-normalize.md](references/process/peak-gain-normalize.md)
- [references/process/audio-concat.md](references/process/audio-concat.md)
- [references/process/delivery-variants.md](references/process/delivery-variants.md)
- [references/process/still-image-youtube-video.md](references/process/still-image-youtube-video.md)
- [references/process/youtube-manual-upload.md](references/process/youtube-manual-upload.md)

The links above are the task-specific entry points. If another reference is
needed, inspect or search the relevant `references/` directory.

For the reasoning behind the current YouTube output defaults, see
[references/decisions/youtube-output-policy.md](references/decisions/youtube-output-policy.md).

## Prerequisites

Running generated commands requires the `ffmpeg` CLI to be available in the
user's shell. `ffmpeg -version` is mandatory before executing any workflow
command on the user's machine:

```sh
ffmpeg -version
```

If `ffmpeg -version` fails or `ffmpeg` is unavailable, stop command execution
and ask the user to install or expose `ffmpeg` in `PATH`.

## First-Cut Guardrails

Unless the user explicitly changes the first-cut workflow:

- Use copied ZOOM H4essential recording folders as the source.
- Assume orchestra or ensemble rehearsal recording.
- Treat trimming as expected and usually required.
- Use a two-phase simple-gain route: measurement first, finishing second.
- Use `loudnorm=print_format=json` only for measurement JSON.
- Use `volume=...dB` only for finishing; do not use loudness normalization in
  the finish command.
- Target peak is `-0.5 dBTP`.
- Do not use compression.
- Inspect the selected source's sample rate, PCM format, and channels with
  `ffprobe`. For H4essential PCM 32-bit float input, keep the source sample rate
  and `pcm_f32le` through trimming and the adjusted WAV master by default.
- Create delivery files directly from that master. Convert to 48 kHz when
  encoding the YouTube AAC audio; avoid a 192 kHz or 44.1 kHz intermediate.
- Create hi-res or lo-res PCM copies only when requested, following
  [delivery-variants.md](references/process/delivery-variants.md).
- For multiple WAVs, trim and gain-adjust each selected WAV first, then
  concatenate.
- Retain the selected adjusted WAV as a separately named master; do not replace
  it with a delivery format.
- When the user asks for mobile playback, lightweight sharing, or a
  Drive-friendly file, create a requested delivery variant after preserving the
  master.
- End at a YouTube-uploadable video file.
- Verify the final file's duration, audio format, reported bitrate, and decoded
  true peak. The WAV master's peak alone does not verify the AAC output.
- Upload manually through the YouTube Studio Web UI.
- Do not introduce YouTube API upload, OAuth, scheduled publishing, or metadata
  automation unless the user explicitly asks to expand the skill.

## Response Style

From igapyon/igapyon-agent-skills, Apache-2.0.

More from igapyon/igapyon-agent-skills