Init Elektra

/init-elektra

Run or re-run Elektra's First Session Protocol.

When to use

• First session in a new project (runs automatically via session-init hook)
• User wants to re-onboard: changed role, new team member, different project context
• Reset after cloning into a new machine or workspace

Behavior

Mode detection

Check for existing onboarding state:

if MEMORY.md exists AND contains user_profile:
  MODE = "reset"
else:
  MODE = "fresh"

Fresh mode (no prior onboarding)

Run the full First Session Protocol from CLAUDE.md:

1. Phase A: Self-Onboard -- read README, package.json/pyproject.toml, git history, directory structure. Save to project_discovery.md.
2. Phase B: Onboard User -- ask all questions in TWO batched prompts (role, experience band, team context, current work, git workflow, testing philosophy, deployment target). Save to user_profile.md.
3. Phase C: Initialize Memory -- create MEMORY.md with index entries.
4. Phase D: Configure Standing Orders -- adapt behavior based on discovered context.
5. Phase E: Dispatch -- classify user's current task and route to the right Standing Order.

Reset mode (re-onboarding)

1. Inform the user that existing onboarding data was found.
2. Ask: "Re-run full onboarding (project + user), or just update your user profile?"
   • Full reset: archive existing project_discovery.md and user_profile.md (rename with _prev_YYYYMMDD_HHMMSS timestamp suffix), then run Phases A-E.
   • Profile only: archive user_profile.md (timestamp suffix), re-run Phase B only, update MEMORY.md index.
3. After re-onboarding completes, confirm the new register (S1-S4) and announce readiness.

Rules

• Never delete memory files outright -- archive with _prev_YYYYMMDD_HHMMSS timestamp suffix so nothing is lost.
• Always run Phase A before Phase B in full reset -- project context informs which Phase B questions to skip.
• Batch all onboarding questions in TWO prompts, never one-by-one.
• After completing onboarding, announce: [ELEKTRA] Onboarding complete. Register: {band}. Ready.