Skip to content

Project Setup Guide

How projects interact with BoB — from initialization to ongoing maintenance.

Run cdi (universal items) and cdprov (manifested registry items) in the project you want to connect to BoB:

Terminal window
cd ~/Sites/my-project
cdi # Symlinks universal commands, agents, hooks
cdprov # Provisions registry items per the manifest
# Or pass an explicit path: cdi /path/to/project

cdi is the wiring step — it creates .claude/ and the universal symlinks. cdprov reads BOB_SOURCE/provisions/<project>.json (falling back to _default.json) and adds symlinks for every registry skill, command, and agent the project declares. The two are split because universal items don’t need a manifest, but registry items do.

.claude/
├── commands/ ← Symlinks to BOB_SOURCE/commands/
├── hooks/ ← Symlinks to BOB_SOURCE/hooks/
├── agents/ ← Symlinks to BOB_SOURCE/agents/
└── settings.json ← Hook configuration + project metadata

Workflow agents (solution-architect, project-historian, lessons-extractor) are always included.

Running cdi on an existing project is non-destructive:

  • Existing symlinks are skipped
  • Local files are preserved
  • Only missing components are added

2. What Gets Linked vs What’s Auto-Available

Section titled “2. What Gets Linked vs What’s Auto-Available”
Component Symlinks Needed? Why
Skills No Auto-loaded from ~/.claude/skills/ (deployed from BOB_SOURCE)
Commands Yes Must exist in project’s .claude/commands/
Agents Yes Must exist in project’s .claude/agents/
Hooks Yes Referenced by path in project’s settings.json
Scripts No Referenced by absolute paths

Skills are the exception. Drop a skill into BOB_SOURCE/skills/, deploy, and every project can use it immediately — no symlinks, no linking step.

Everything else needs a symlink from the project’s .claude/ directory pointing back to the source in BOB_SOURCE.

Pull the latest source and deploy:

Terminal window
cd ~/projects/bigbrain && git pull
bash scripts/deploy.sh

Existing symlinks auto-propagate. Project symlinks point to BOB_SOURCE, so changes there are picked up instantly. The deploy step syncs skills and other auto-loaded content to ~/.claude/.

New items need linking. If BoB gains a new command or agent that your project doesn’t have yet, you’ll need to link it:

Terminal window
cd ~/Sites/my-project
cdl commands new-command
cdl agents new-agent

Or grab everything at once:

Terminal window
cdl --all

Use cdl to link individual items from BoB into your current project:

Terminal window
# Link a command
cdl commands start-work
# Link an agent
cdl agents drupal-expert
# Link a hook
cdl hooks check-branch.sh
# See what's available
cdl --list
cdl --list agents
# Link ALL missing global items
cdl --all
  • All global commands
  • All global hooks
  • Workflow agents (solution-architect, project-historian, lessons-extractor)
  • Domain expert agents — link individually as needed (cdl agents drupal-expert)
  • Skills — they don’t need symlinks (already auto-available)
Symbol Meaning
Successfully linked
Already linked (skipped)
! Local file exists (not overwritten)
Global item not found

When you develop something useful in a project and want to share it globally, use cdp:

Terminal window
cdp .claude/commands/my-command.md
cdp .claude/agents/my-agent.md
cdp .claude/hooks/my-hook.sh
cdp .claude/skills/my-skill/ # Promotes entire directory
  1. Copies the file/directory to BOB_SOURCE/<type>/
  2. Removes the local file
  3. Creates a symlink pointing to the new global copy in BOB_SOURCE

The safest workflow is develop → test → promote:

  1. Develop a new tool in a low-risk project (e.g., bodmail)
  2. Test it thoroughly in that project
  3. Promote to BoB: cdp .claude/commands/my-tool.md
  4. Link into other projects: cdl commands my-tool
Path Pattern Promotes To
commands/*.md BOB_SOURCE/commands/
agents/*.md BOB_SOURCE/agents/
hooks/*.sh BOB_SOURCE/hooks/
skills/*/ BOB_SOURCE/skills/ (auto-available after deploy)

Use cdb to see what’s linked and what’s missing:

Terminal window
# Full dashboard (global inventory + project summary)
cdb
# Matrix view: every global item vs every project
cdb --matrix
# Detailed check for a specific project
cdb --check ~/Sites/my-project
# Just global inventory
cdb --global-only
# Just project summaries
cdb --projects-only
Global Item bodmail nanawall
───────────────────────────── ──────────── ────────────
Commands
journal ● ●
Hooks
check-branch.sh ● ●
strip-ai-boilerplate.sh ✗ ✗
Symbol Meaning
green Symlinked to global
blue Local file (same name as global, but not linked)
red Missing from project
yellow Local-only (exists in project, not in global)

The matrix also shows Local-Only Items — things in your project that don’t exist in BoB. These are candidates for promotion via cdp.

Symlinks can break if:

  • You moved the project directory
  • You re-cloned the project on a different machine
  • ~/.claude/ was re-deployed from BOB_SOURCE

Fix it with:

Terminal window
cd ~/Sites/my-project
cdl --all

This re-creates all standard symlinks. For domain expert agents, link them individually:

Terminal window
cdl agents drupal-expert
cdl agents cloudflare-expert

To verify everything is healthy:

Terminal window
cdb --check ~/Sites/my-project

Set up the IaC dev lifecycle so the project’s dev environment is always testable:

Terminal window
# Copy the dev manifest template (from BOB_SOURCE or deployed BOB_HOME)
cp "$BOB_SOURCE/templates/dev.json" ./dev.json
# Customize: edit server command, port, health endpoint, auth adapter
vi dev.json
# Set up seed data
mkdir -p seed/
cp "$BOB_SOURCE/templates/seed/users.json" ./seed/
cp "$BOB_SOURCE/templates/seed/001-base-data.sql" ./seed/
cp "$BOB_SOURCE/templates/seed/002-sample-entities.sql" ./seed/
# Customize the SQL for your schema
# Test it
dev-up . --verbose

This gives you:

  • /dev-up — bring the dev environment to fully testable state on demand
  • Warp-drive auto-manages dev health during autonomous coding
  • Standardized test users across all projects (admin@test.local / admin123)
  • Seed data covering all entity lifecycle states

See dev-lifecycle.md for the full guide.

Once a project is initialized, use the product hierarchy to plan and track work:

/vision Create docs/vision.md — product vision & strategy
/capability Create a capability (epic) as a GitHub Issue
/requirement Create a requirement with acceptance criteria
/warp-drive Autonomous coding loop (picks up approved requirements)

Start with /vision to establish why the product exists, who it’s for, and where it’s going. Then break the vision into capabilities, capabilities into requirements, and let warp-drive handle the rest.