soundcue-audio-review

name	soundcue-audio-review
description	Review podcast or meeting audio with transcript-linked notes, create suggested edits, and maintain local linear audio versions.

SoundCue Audio Review

Use this skill when the user wants to annotate a podcast, meeting recording, or generated audio segment and ask Codex to produce a local suggested edit.

Core Boundary

• One Codex conversation maps to one SoundCue project.
• The UI owns selection, comment, preview, and review state.
• Codex owns local file operations and provider calls through typed tools.
• Audio algorithms are adapter-based: local ffmpeg / STT / TTS commands can be configured, and cloud providers can be added later.
• Never modify or overwrite the original audio version. A confirmed edit creates a new derived vN audio version.
• Never treat a suggested edit as created until the user explicitly confirms the new version.

Workflow

1. Read the current project with read_audio_patch_project.
2. If needed, import an audio file with import_audio_file or create a text-first draft with create_audio_project_from_text.
3. Read the annotation selection and intent.
4. Create a suggested edit with generate_audio_patch_candidate.
5. Ask the user to review before/after in the UI.
6. Create the derived version only after confirmation with apply_audio_patch; the base version must remain unchanged.

Data Model

• Project: bound to threadId, contains assets, versions, annotations, patches.
• Version: linear v1, v2, v3, never a Git-like branch.
• Annotation: user intent attached to transcript/audio range.
• Patch: internal object for a generated suggested edit with preview audio, transcript change, and edit summary.

Keep the user-facing explanation in terms of transcript selection, notes, suggested edits, and local versions.