prep-interview

Workflow

1. Resolve {user_dir} and gate

Application data uses the split layout defined in references/tracker-schema.md — read it before any read or write.

Require CareerNavigator/profile.md, ExperienceLibrary.json, and tracker.json. If missing:

Run /career-navigator:launch first to initialize Career Navigator.

Story corpus gate:

• Preferred source: CareerNavigator/StoryCorpus.json.
• If missing, run or suggest mine-stories first.
• If user wants to continue immediately, proceed with ExperienceLibrary.json fallback and clearly label lower story-specific confidence.

2. Identify target

From user message or tracker:

• Prefer application_id if given; else match company + role against tracker.json applications[]. If an application_id matches no row's id, check each row's previous_ids before declaring no match — prep briefs saved under earlier ids still resolve.
• Keep the matched row's detail_file, contacts_file, and application label; the write in step 6 needs them.
• Pull prior stage history and notes from the row's detail_file (CareerNavigator/applications/<application_id>.json) — tracker.json holds only latest_stage and latest_stage_date. Load the contacts_file only if interviewer background is needed, filtering its contacts[] to entries whose application equals the row's application.
• If ambiguous, ask one short disambiguation question.

Collect if missing:

• interview_stage: recruiter | hiring_manager | technical | panel | executive | final (default: infer from wording; recruiter call / phone screen → recruiter).
• Optional: interviewer name/title, interview date, JD text or paste.

3. Optional STT / voice

If the mcp-voice MCP exposes speak / listen, use them per agents/interview-coach/AGENT.md. If the user is speaking or pastes a transcript from host STT, pass it as user_audio_transcript in the handoff to interview-coach.