Skip to content

Backup & Recovery

The BoB source repo (BOB_SOURCE, default ~/projects/bigbrain) is hosted on GitHub (paulirv/bigbrain). All critical state is version-controlled and deployed to ~/.claude/ (BOB_HOME) at runtime:

  • Skills, commands, agentsskills/, commands/, agents/, registry/
  • Provision manifestsprovisions/*.json (one per project)
  • Scripts and binariesscripts/, bin/
  • Shared settingssettings.json
  • Templates and hookstemplates/, hooks/
  • Documentationdocs/, runbooks/

These are machine-specific or transient and are gitignored:

Item Why Recovery
settings.local.json Machine-specific (automation level, RDB) Re-set via /automation and /rdb
.mcp.json Machine-specific MCP server paths Re-configure MCP servers
CLAUDE.local.md Machine-specific overrides Re-create if needed
projects/ Claude Code session metadata Regenerated automatically
history.jsonl Conversation history Not recoverable (ephemeral)
token-usage.jsonl Token usage logs Not recoverable (analytics only)
sessions/, remote/ Runtime state Regenerated on use
.credentials.json Sensitive auth tokens Re-authenticate

Terminal only. Every command in this guide is a terminal command, not a slash command — recovery runs from a shell, often on a fresh machine before any Claude Code session exists, so there is nothing to type /-commands into. The one exception worth knowing: where cdprov re-provisions a project below, the in-session equivalent is /provision — but it presupposes a working session, which recovery is what restores.

Full Recovery (Fresh Machine or Total Loss)

Section titled “Full Recovery (Fresh Machine or Total Loss)”

Terminal:

Terminal window
# 1. Clone the BoB source repo
git clone git@github.com:paulirv/bigbrain.git ~/projects/bigbrain
# 2. Set environment variables (in ~/.zshrc)
export BOB_HOME="$HOME/.claude"
export BOB_SOURCE="$HOME/projects/bigbrain"
# 3. Deploy to ~/.claude/ and install CLI aliases
bash ~/projects/bigbrain/scripts/deploy.sh
bash ~/projects/bigbrain/scripts/bob-install.sh
source ~/.zshrc
# 4. Set machine-specific config
# (automation level, RDB, etc. — these are interactive)
# 5. Re-provision all projects
# For each project directory:
cd ~/Sites/project-name
cdi # Initialize global tooling symlinks
cdprov # Re-apply provision manifest

Use the cdr command (claude disaster recovery) to automate steps 1-4. cdr is terminal only — it has no slash form:

Terminal:

Terminal window
cdr status # Show current tooling and warp-drive state
cdr check # Health check — verify critical files exist
cdr recover [project] # Recover a project: fetch + re-run cdi
cdr reprovision # Re-provision ALL known projects from manifests
cdr recover-global # Full recovery: clone/pull + aliases + reprovision
cdr reset [project] # Emergency: remove warp-drive state file

The cdr alias is installed by bob-install.sh and points to scripts/lib/cdr.sh.

If specific files are corrupted but the source repo is intact:

Terminal:

Terminal window
# In the source repo (BOB_SOURCE)
cd ~/projects/bigbrain
# Reset corrupted files from remote
git fetch origin master
git checkout origin/master -- path/to/corrupted/file
# Or reset everything to remote state
git reset --hard origin/master
# Re-deploy to runtime
bash scripts/deploy.sh

Terminal:

Terminal window
cd ~/Sites/project-name
cdi # Re-link global items
cdprov # Re-apply project manifest

The cdprov step has an in-session equivalent — /provision — once a Claude Code session is running; cdi is terminal only.

  1. Commit frequently — All changes to skills, commands, agents, provisions should be committed in BOB_SOURCE
  2. Push to remote — After committing, push to GitHub: cd ~/projects/bigbrain && git push
  3. Deploy after changes — Run scripts/deploy.sh to sync to ~/.claude/
  4. Health check — Run cdr check periodically or let the session-start check catch issues
  5. Don’t manually edit — Use BoB CLI commands (cdi, cdprov, cdp, cdl) which keep state consistent