Brainerd Init

Initialize Brainerd in the current repo and route to the correct harness-specific setup path.

Detect the harness first

1. If the active session exposes Pi Brainerd tools, or you are in Pi, use the guarded /brainerd-init command surface. Do not use the shell wrappers in that branch.
2. If CODEX_THREAD_ID is present, use the Codex wrapper:

./scripts/brainerd-codex.sh init

3. If BRAINERD_CLAUDE_SESSION_ID or BRAINERD_CLAUDE_TRANSCRIPT_PATH is present, use the Claude wrapper:

./scripts/brainerd-claude.sh init

4. If none of those signals match, stop and report that harness detection failed. Do not guess.

Workflow

1. Run the matching init path for the detected harness.
2. Review the bootstrap preview before writing it.
3. Apply the optional bootstrap note only when the user explicitly wants it.
4. Keep edits limited to brain/ and the managed Brainerd block in AGENTS.md.
5. On the Claude branch, keep CLAUDE.md as a thin shim and preserve Claude imports under brain/imports/claude/.

Guardrails

• Do not edit arbitrary repo files outside brain/ and the managed Brainerd block in AGENTS.md.
• Stop if the managed Brainerd block is duplicated or malformed.
• End with a short summary of what was created and whether bootstrap was previewed or applied.