Bifrost Setup

Full guided setup — from zero to working memory system. Walk the user through each step, adapting based on what they already have.

/setup

Step 1: Check Dependencies

Run these checks via Bash:

• git --version — required. If missing, stop and tell the user to install Git.
• python3 --version — required for hooks. If missing, warn and link to python.org.

Step 2: Check for Existing Config

Look for ~/.config/bifrost/config. If it exists, show current values and ask if they want to reconfigure or start fresh.

Also check for legacy configs (~/.config/asgard/config, ~/.config/munin/config). If found, offer to migrate values.

Step 3: Memory Repo

Ask: "Do you have an existing memory repo, or should we create one?"

If they have one:

• Ask for the path. Validate it exists and is a git repo.
• Check for expected structure: MEMORY.md, inbox/, journal/, procedures/. If any are missing, offer to create them.

If they need a new one:

• Ask where to create it (e.g., ~/projects/my-memory).
• Create the directory and initialize git: git init
• Scaffold the full directory structure:

  MEMORY.md
  procedures/procedures.md
  journal/
  inbox/
  context-trees/
  .gitignore
  

• Seed MEMORY.md with a starter template:

  # Memory
  
  ## Identity
  - Name:
  - Role:
  
  ## Preferences
  
  ## Active Projects
  
  ## Technical Environment
  

• Seed procedures/procedures.md:

  # Procedures
  
  Index of learned workflows. Individual procedure files live in this directory.
  

• Seed .gitignore:

  .DS_Store
  

• Ask if they want to add a git remote (for syncing across machines). If yes, ask for the URL, run git remote add origin <url>.
• Make an initial commit: git add -A && git commit -m "Initial memory repo"
• If remote was added, push: git push -u origin main
• Tell the user: "Your memory repo is ready. Seed MEMORY.md with what your agents should know about you — identity, preferences, projects, technical environment."

Step 4: Machine Name

Ask for a machine name. This identifies the machine in inbox filenames. Suggest the hostname as default: run hostname via Bash. The name should be short, lowercase, hyphenated (e.g., personal-laptop, home-desktop). Validate: only allow [a-z0-9-] — reject names with spaces, slashes, or special characters.

Step 5: Rules Scope

Ask where to plant the rules file (AskUserQuestion, single-select):

• Global (recommended) — ~/.claude/rules/bifrost.md — active in all projects
• Project only — .claude/rules/bifrost.md — active only in current project

Step 6: Write Config File

mkdir -p ~/.config/bifrost
cat > ~/.config/bifrost/config << EOF
BIFROST_REPO=<path>
BIFROST_MACHINE=<name>
BIFROST_JOURNAL_DAYS=2
EOF

Step 7: Write Rules File

Write the rules file to the chosen location. Critical: Do NOT hardcode the repo path. Instead, instruct Claude to read the config file at runtime:

# Bifrost: Memory Awareness

You have persistent memory. To find its location, read `~/.config/bifrost/config`
and use the `BIFROST_REPO` value.

## Session Context

Your core memory (MEMORY.md) and recent journal entries are already in your
context — loaded by the bootstrap hook at session start. Don't re-read them.
Use Read/Grep on the layers below when you need specific workflows or
historical context that wasn't loaded at startup:

| Layer | Path (relative to BIFROST_REPO) | Use |
|-------|------|-----|
| Core memory | `MEMORY.md` | Stable facts — already loaded as context |
| Procedures | `procedures/` | Learned workflows — Read when doing multi-step tasks |
| Journal | `journal/` | Recent events and decisions — Grep for temporal context |
| Context trees | `context-trees/` | Project status and relationships |

## When to Recall

Before planning or making decisions that could benefit from historical context:
- Use Grep on `journal/` and `procedures/` for specific topics
- For deep searches across all layers, use `/odin <topic>`

## Corrections

When the user corrects you about something memory should know, note the
correction explicitly: "Noted correction: <what changed>." This flags it
for future consolidation.

## Constraints

- Memory files are **read-only during sessions** — never edit them directly
- Don't recite memory to the user — use it silently to inform your behavior

Step 8: Verification

Confirm everything is wired up:

• Config file exists and has correct values
• Rules file exists at chosen path
• Memory repo is accessible and has MEMORY.md
• Git remote is configured (warn if not — syncing won't work without one)
• Report: "Bifrost is configured. Memory will load at session start and transcripts will be captured at session end."