Hooks

brainjar integrates with the active platform’s hook system for automatic context injection. When hooks are installed, brainjar syncs your config on every session start — no manual brainjar sync needed. Claude Code, Codex, and Cursor all expose a hook surface; the active context’s platform decides where the hook is written.

See CLI reference for full flag details.

Installing hooks

brainjar hooks install                  # Default scope: project (shared with collaborators)
brainjar hooks install --scope local    # Per-checkout, not shared
brainjar hooks install --scope user     # Global across every project

The default project scope writes the hook into the repo’s shared settings (e.g. Claude Code’s .claude/settings.json) so every collaborator picks it up on clone. Use local for a per-checkout override that stays out of version control (Claude Code: under projects.<root>.hooks in ~/.claude.json), or user for a personal global default. Pass --project <path> to target a different repo root; ignored when --scope=user.

Scope availability varies by platform — brainjar platform list shows which scopes each adapter supports.

How it works

The hook runs brainjar sync on the platform’s session-start event. This means:

1. You change a soul, persona, procedure, rule, or skill (or load a different brain)
2. Next time the agent starts a session, the hook fires
3. The platform’s managed config block (e.g. CLAUDE.md, AGENTS.md) is rewritten with the latest composed prompt and skills are emitted to disk
4. The agent picks up the new behavior

No manual sync step needed. The hook is idempotent — it only rewrites the brainjar-managed section between the <!-- brainjar:begin --> / <!-- brainjar:end --> markers.

Managing hooks

brainjar hooks status    # Show whether the hook is registered, and at which scope
brainjar hooks remove    # Remove the hook from settings

Manual sync

If you’re not using hooks, sync manually after making changes:

brainjar sync