Skip to content

init.md

Latest commit

 

History

History
38 lines (27 loc) · 1.84 KB

init.md

File metadata and controls

38 lines (27 loc) · 1.84 KB

rekal init

Role: Bootstrap Rekal in a git repository. The only command a developer must run once per repo.

Invocation: rekal init.

Preconditions

  • Must be run inside a git repository. Otherwise exit with "not a git repository".

What init does

  1. Resolve git root — Exit if not in a git repo.
  2. Check if already initialized — If .rekal/ exists, print "already initialized" and exit. User must run rekal clean first to reinitialize.
  3. Create .rekal/ — Directory for local databases.
  4. Create data DB — Open .rekal/data.db, run data DDL (sessions, turns, tool_calls, checkpoints, files_touched, checkpoint_sessions, checkpoint_state).
  5. Create index DB — Open .rekal/index.db, run index DDL (turns_ft, tool_calls_index, files_index, session_facets, file_cooccurrence, session_embeddings, index_state).
  6. Update .gitignore — Append .rekal/ if not already present.
  7. Install hooks:
    • post-commit — runs rekal checkpoint
    • pre-push — runs rekal push
    • Hooks contain the marker # managed by rekal. Existing non-Rekal hooks are not overwritten.
  8. Create orphan branch — rekal/<email> with empty rekal.body and dict.bin. If the branch exists on the remote, fetch it. If it exists locally, leave it.
  9. Import existing data — If the orphan branch has data (body > 9 bytes), import sessions and checkpoints into data DB.
  10. Install Claude Code skill — Write .claude/skills/rekal/SKILL.md for agent integration.
  11. Gitignore .claude — If .claude/ already existed (user has settings, CLAUDE.md, etc.), only ignore .claude/skills/. Otherwise ignore the entire .claude/ directory.
  12. Initial checkpoint — Capture any existing sessions.
  13. Print — Rekal initialized.

No flags

No user-facing flags. Non-interactive.