| name | interview-capture |
| description | Opt-in post-interview capture: transcribe the user's own audio (user audio only), extract structured takeaways, and update tracker.json. Surfaces employer policy warning once; jurisdiction/retention per spec §13.1. Uses the local mcp-voice MCP (faster-whisper STT) when available. Not an agent—a skill orchestrating tools and track-application patterns. Also invocable via /career-navigator:interview-capture.
|
| triggers | ["/interview-capture","/career-navigator:interview-capture","capture my interview","transcribe my interview notes","log interview from audio","interview audio capture"] |
Scope (§13.1)
- User audio only. Do not capture or transcribe the employer’s side without appropriate consent and product scope—that remains out of scope.
- Opt-in: Run this skill only when the user explicitly asks to capture/log from audio or runs
/career-navigator:interview-capture.
- Employer policy warning (once): Some employers prohibit any recording of interviews. Before the first capture session, show the warning from spec §13.1 and require acknowledgment. Persist to
{user_dir}/CareerNavigator/interview-capture-settings.json:
{
"employer_warning_acknowledged": true,
"acknowledged_at": "YYYY-MM-DD",
"user_opted_in": true
}
If employer_warning_acknowledged is already true, do not repeat the full warning—one line reminder max.
- Retention / jurisdiction: Use
profile.md location for GDPR/CCPA framing per spec. If unclear, ask once which retention stance applies and store under interview-capture-settings.json as retention_note.
Workflow
1. Resolve {user_dir} and gate
Require CareerNavigator/profile.md and tracker.json. If missing, send user to /career-navigator:launch.
2. Opt-in and warning
- If
interview-capture-settings.json does not exist or employer_warning_acknowledged is false: print the employer policy warning; on user acknowledgment, write the settings file.
- If user has not opted in to capture for this session, confirm before processing audio.
3. Match application
Identify application_id (or company + role) for the interview being logged. If unknown, ask before writing tracker updates.
4. Transcription (STT)
- Discover tools: if the
mcp-voice MCP exposes listen, invoke it to record the user speaking their recap aloud (prompt them to speak naturally). If the user has a pre-recorded audio file, ask for its path and note that listen records live — for file-based transcription, the user should paste the transcript manually.
- If no STT tools: fall back to user-pasted transcript or
/career-navigator:interview-debrief when shipped; do not invent transcripts.
5. Structure and write tracker
From the transcript (user side only), extract:
- Overall tone / how it went
- Stage, participants (from user statements)
- Topics, surprises, red flags / positive signals
- Committed follow-ups
- Next-round or outcome hints
Append to applications[].notes with prefix [capture] and date; update stage_history or next_step via the same patterns as track-application when appropriate.
6. Artifacts (optional)
Save transcript text to {user_dir}/CareerNavigator/interview-capture/ as {application_id}-{YYYY-MM-DD}-transcript.txt if the user wants a file copy.
Guardrails
- Honest over encouraging when summarizing performance.
- No multi-party recording claims; label transcript as user channel only when other speakers are quoted from user recall.
