Soul Audit — Agent Identity Builder (re-run / deepen surface)

Update the agent's identity through an interactive interview, any time after bootstrap. This skill does NOT generate files from scratch and does NOT write identity prose itself: it records the user's answers into the shared answer bank, then lets the bootstrap render engine produce the files. That keeps one source of truth — state/interview.json — no matter which surface (first-boot bootstrap, this skill, a future re-audit) collected the answers.

IMPORTANT: Content comes from the USER'S OWN ANSWERS, recorded verbatim. NEVER invent, embellish, or default an identity answer. The render gate structurally refuses until required answers exist and were read back and confirmed by the human.

Contract

This skill guarantees:

• Answers land in the shared bank via gbrain bootstrap interview --set KEY "value" (verbatim, one key per answer; --skip KEY records an explicit decline).
• The read-back ritual runs before any render: --show the full answer set, mirror it to the user, then --confirm <hash> with the hash --status printed. Confirming a set the user never saw fails by design.
• Files re-render through the ONE render path: gbrain bootstrap render --only SOUL.md --force (repeat per file). --force backs the old file up under .gbrain-bootstrap-backups/<ts>/ first — nothing is clobbered without a recoverable copy.
• Each phase is independent and re-runnable; updating one answer and re-rendering one file is a complete, valid run.

The command loop

gbrain bootstrap interview --show            # what the bank holds now (start here)
gbrain bootstrap interview --set KEY "..."   # record one updated answer, verbatim
gbrain bootstrap interview --skip KEY        # explicit decline for an optional key
gbrain bootstrap interview --status          # gate + read-back hash
gbrain bootstrap interview --confirm <hash>  # after reading answers back to the user
gbrain bootstrap render --only SOUL.md --force

Render targets: SOUL.md, USER.md, ACCESS_POLICY.md, HEARTBEAT.md, AGENTS.md (re-render whichever files the changed keys feed).

Phases (each independent, each re-runnable)

Phase 1: Identity Interview

Ask: "What is this agent to you? Research partner? Executive assistant? Thinking partner? All of the above?" Keys: AGENT_NAME, AGENT_PURPOSE, SOUL_RELATIONSHIP. Renders: SOUL.md.

Phase 2: Vibe Calibration

Show 3-4 communication style examples:

• Formal: "I've prepared a comprehensive analysis of the situation..."
• Direct: "Here's what's happening. Three things matter."
• Technical: "The root cause is in the connection pooling. Here's the fix."
• Casual: "Yeah so basically the thing is broken because X. Easy fix." Ask which feels right; also ask for a reply that made them wince and why. Keys: VOICE_REGISTER, SOUL_WINCE, SOUL_MODE_DEFAULT. Renders: SOUL.md.

Phase 3: Mission Mapping

Ask: "What are your top 3-5 goals? What are you trying to accomplish?" Keys: AGENT_TOP_JOBS, SOUL_GOOD_OUTPUT, SOUL_WORLDVIEW. Renders: SOUL.md, AGENTS.md.

Phase 4: User Profile

Ask: "Tell me about yourself. What do you do? What are you working on? Who are the key people in your world?" Keys: PRINCIPAL_NAME, PRINCIPAL_CONTEXT, PRINCIPAL_TIMEZONE. Renders: USER.md.

Phase 5: Boundaries

Ask: "Who should have access to your brain? Are there people who should see some but not all? Anyone to keep out entirely?" Key: ACCESS_TIERS (and review the standing-disclosure sections). Renders: ACCESS_POLICY.md.

Phase 6: Operational Cadence

Ask: "How often should the agent check in? Morning briefing? End of day summary? What recurring jobs do you want?" Keys: HEARTBEAT_CADENCE, QUIET_HOURS. Renders: HEARTBEAT.md. Jobs ship disabled — enable one at a time after a manual dry run (the ritual is in the rendered file).

Run gbrain bootstrap interview --init first if the bank does not exist yet (a workspace that never ran bootstrap); it prints the full question list.

Output Format

Report exactly what changed: which keys were updated, which files re-rendered, and where the backups landed. Example: "Soul audit: updated VOICE_REGISTER + SOUL_WINCE, re-rendered SOUL.md (backup in .gbrain-bootstrap-backups/...). Re-run any phase anytime."

When it fails

Follow the agent operator protocol for any gbrain error code, exit code, [AGENT] block or notice block. Specific to this skill:

• gbrain bootstrap interview --confirm <hash> fails: the hash must come from a read-back the user actually saw. Re-run gbrain bootstrap interview --status, read the answers back, and confirm only with the user's agreement.
• Rendering refuses because required answers are missing: ask the user for them; never fill identity answers yourself.

Anti-Patterns

• Writing identity prose yourself instead of recording the user's words into the bank (the render engine is the only author of these files)
• Hand-editing SOUL.md/USER.md directly — the next render overwrites it; edit the ANSWER, then re-render
• Confirming a read-back hash for answers the user never saw
• Making the audit mandatory or asking all 6 phases in one go (overwhelming; each is independent — momentum beats completeness)
• Shipping pre-filled identity content (privacy violation)
• Not offering to re-run individual phases