Menu
Handle voice message processing. Use internally when the router encounters a voice-type message. Do NOT invoke directly.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Robust, line-based file editing. Replace a range of lines in a file with new content by line number instead of exact string matching. Use when the user says 'robust-edit', '/robust-edit', 'replace lines', 'edit lines N to M', or when character-level matching with the Edit tool keeps failing on multi-line changes or uncertain whitespace.
Review code changes for quality, security, and correctness. Use when the user says "review this PR", "review these changes", "check my code", "look at what I changed", or after implementing a feature. Produces a severity-organized report.
Diagnose and fix context window issues. Use when Claude seems confused, forgetful, slow, or the user says "context is full", "you forgot", "you're repeating yourself", "start fresh", or "why are you ignoring my instructions".
Structured git workflow for commits, branches, and PRs. Use when the user says "commit", "create a PR", "new branch", "prepare for review", "ship this", or "push my changes". Safe, conventional git operations with checkpoints.
Deep interview to reach shared understanding before building. Use when starting a complex feature, when requirements are unclear, or when the user says "grill me", "interview me", "ask me questions", "let's figure this out", or "grill me lightly" for a quick version.
Approval gate for dangerous operations. Use internally before executing destructive commands, file deletions, git pushes, or external API calls from the web chat. Do NOT invoke directly.
| name | oc-voice |
| description | Handle voice message processing. Use internally when the router encounters a voice-type message. Do NOT invoke directly. |
| user-invocable | false |
| allowed-tools | Read Bash(python*) |
Handles voice messages received through the web chat. By the time you see them, the audio has already been transcribed by Whisper.
The web UI records audio → uploads to /api/voice → Python server transcribes with whisper --model turbo → queues the transcribed text as type voice.
The message in the queue looks like:
{
"type": "voice",
"content": "hey can you check if the tests pass",
"metadata": "{\"audio_file\": \"voice_abc123.webm\", \"language\": \"en\"}"
}
content IS the user's message — process it like any text messageGood transcription:
Bad transcription (ask user to repeat):
The original audio is saved at data/uploads/<audio_file> if you need to reference it. You cannot re-transcribe it, but you can note its existence.