setup-project

name	setup-project
description	Bootstraps a new project directory with base configuration files, MCP servers, API connections, and template skills from OutworkOS. Use when setting up a new project, initializing a workspace, or configuring MCP integrations for a new directory.
argument-hint	<project-name>

/setup-project

Bootstrap a new project directory under $OUTWORKOS_PARENT with base configuration files from $OUTWORKOS_ROOT (the OutworkOS repo — source of truth for shared skills and credentials).

Key concepts:

• $OUTWORKOS_ROOT — the OutworkOS repo (where you run Claude from, contains scripts/, skills/, .mcp.json)
• $OUTWORKOS_HOME — data directory (e.g., ~/.outworkos)
• $OUTWORKOS_PARENT — where project folders live (e.g., ~/.outworkos/projects/)
• These CAN be the same directory if the user configures it that way

Trigger

User invokes /setup-project or /setup-project <project-name>

Steps

1. Validate Environment & Create Project Directory

1. Confirm $OUTWORKOS_ROOT is set and the template directory exists with these files:

   • $OUTWORKOS_ROOT/.mcp.json
   • $OUTWORKOS_ROOT/skills/context-map/
   • $OUTWORKOS_ROOT/skills/log/
   • $OUTWORKOS_ROOT/skills/todoist/ If any are missing, warn but continue (some may not be needed).

2. Confirm $OUTWORKOS_PARENT is set. If not, default to $OUTWORKOS_HOME/projects or ask the user.

3. Get the project name:

   • If provided as argument, use it
   • Otherwise, ask: "What's the name of this project?"
   • Slugify the name for the directory (lowercase, hyphens for spaces, strip special chars)

4. Create the project directory at $OUTWORKOS_PARENT/<slug>/:

   • If the directory already exists and is not empty, stop with an error: "Project directory already exists. Use a different name or remove it first."
   • If the directory exists but is empty, proceed
   • If it doesn't exist, create it with mkdir -p

5. Set PROJECT_DIR=$OUTWORKOS_PARENT/<slug> for use in all subsequent steps.

Note: The user runs Claude from $OUTWORKOS_ROOT (the repo) to get skills loaded. This step creates the project directory elsewhere — that's expected.

2. Check for Existing Files

Check whether these already exist in the project directory:

• .mcp.json
• .claude/skills/context-map/
• .claude/skills/log/
• .claude/skills/todoist/

For each one that exists, ask the user whether to skip or overwrite it. Use a single AskUserQuestion with multiSelect for efficiency.

3. Verify Vault Access

Verify that outwork init has been run and Vault secrets are accessible. Test by retrieving a known secret:

"$OUTWORKOS_ROOT/scripts/get-secret.sh" todoist_api_token > /dev/null 2>&1 && echo "Vault: OK" || echo "Vault: FAILED — run outwork init first"

If Vault access fails, tell the user to run $OUTWORKOS_ROOT/scripts/outworkos-auth-login.sh first.

4. Link Skills and Hooks

Symlink key directories from $OUTWORKOS_ROOT into the project so skills, hooks, and environment variables work when running Claude from the project directory.

mkdir -p "$PROJECT_DIR/.claude"

# Skills — makes all /commands available
ln -sf "$OUTWORKOS_ROOT/skills" "$PROJECT_DIR/.claude/commands"

# Hooks — SessionStart sets $OUTWORKOS_ROOT, $OUTWORKOS_HOME, $OUTWORKOS_PARENT
ln -sf "$OUTWORKOS_ROOT/.claude/hooks" "$PROJECT_DIR/.claude/hooks"

# Settings — hook config (SessionStart, SessionEnd)
# Copy (not symlink) so the project can customize hooks later
if [ ! -f "$PROJECT_DIR/.claude/settings.json" ]; then
  cp "$OUTWORKOS_ROOT/.claude/settings.json" "$PROJECT_DIR/.claude/settings.json"
fi

This ensures:

• All skills are available (context-map, log, todoist, scan, whats-next, etc.)
• $OUTWORKOS_ROOT, $OUTWORKOS_HOME, $OUTWORKOS_PARENT are set on session start
• Vault scripts (get-secret.sh, set-secret.sh) are reachable via $OUTWORKOS_ROOT/scripts/
• Stays in sync with the repo — no need to copy or update individual files

If .claude/commands or .claude/hooks already exists:

• If it's a symlink pointing to the same target, skip
• If it's a symlink pointing elsewhere, ask the user before replacing
• If it's a regular directory, ask the user whether to replace it with a symlink

5. Supabase Picker (Optional)

Ask the user: "Do you want to add Supabase database connections?"

If no: skip to Step 6 with zero Supabase entries.

If yes:

1. Read SUPABASE_ACCESS_TOKEN from Vault: $("$OUTWORKOS_ROOT/scripts/get-secret.sh" supabase_access_token)
2. Call the Supabase Management API to list projects:

   curl -s -H "Authorization: Bearer <TOKEN>" https://api.supabase.com/v1/projects
   

3. Filter the JSON response to only projects where status is "ACTIVE_HEALTHY"
4. Present a numbered list showing each project's name and ref (the project reference ID)
5. Ask the user to select which projects to add (comma-separated numbers, e.g. "1,3,4")
6. Store the selected projects for Step 6

6. Assemble .mcp.json

1. Read the base .mcp.json from $OUTWORKOS_ROOT/.mcp.json — this contains the 6 base servers (github, netlify, context7, perplexity, fireflies, google-workspace)
2. Parse it as JSON
3. For each Supabase project selected in Step 5, add an entry using the project's name as the key:

   "<project-name>": {
     "command": "npx",
     "args": [
       "-y",
       "@supabase/mcp-server-supabase@latest",
       "--project-ref",
       "<project-ref>"
     ],
     "env": {
       "SUPABASE_ACCESS_TOKEN": "<token-value-from-env>"
     }
   }
   

4. Write the assembled JSON to .mcp.json in the current directory (pretty-printed with 2-space indent)

Important: Use the env field for the access token, NOT --access-token in args. This keeps secrets out of process arguments.

7. Test Connections

After assembling all files, test each API token/key by making a lightweight curl call. Read credentials from Vault via $OUTWORKOS_ROOT/scripts/get-secret.sh and from .mcp.json for MCP-specific tokens.

Run ALL test calls in parallel using separate Bash tool calls for speed.

Tests to run:

Service	Test command	Pass	Fail action
GitHub	curl -s -o /dev/null -w "%{http_code}" -H "Authorization: token <GITHUB_PERSONAL_ACCESS_TOKEN>" https://api.github.com/user	200	Token expired or invalid — regenerate at https://github.com/settings/tokens
Netlify	curl -s -o /dev/null -w "%{http_code}" -H "Authorization: Bearer <NETLIFY_AUTH_TOKEN>" https://api.netlify.com/api/v1/user	200	Token invalid — regenerate at https://app.netlify.com/user/applications#personal-access-tokens
Perplexity	curl -s -o /dev/null -w "%{http_code}" -H "Authorization: Bearer <PERPLEXITY_API_KEY>" -H "Content-Type: application/json" -d '{"model":"sonar","messages":[{"role":"user","content":"ping"}]}' https://api.perplexity.ai/chat/completions	200	API key invalid or expired — check https://www.perplexity.ai/settings/api
Supabase	curl -s -o /dev/null -w "%{http_code}" -H "Authorization: Bearer <SUPABASE_ACCESS_TOKEN>" https://api.supabase.com/v1/projects	200	PAT invalid — regenerate at https://supabase.com/dashboard/account/tokens
Fireflies	curl -s -o /dev/null -w "%{http_code}" -H "Content-Type: application/json" -H "Authorization: Bearer <FIREFLIES_TOKEN>" -d '{"query":"{ user { name } }"}' https://api.fireflies.ai/graphql	200	Token invalid — check https://app.fireflies.ai/integrations
Context7	MCP call: mcp__context7__resolve-library-id (see below)	Returns results	API key invalid — check with Upstash

How to extract each credential (use these exact methods):

1. Read .mcp.json with the Read tool and parse as JSON. Extract:

   • GITHUB_PERSONAL_ACCESS_TOKEN → mcpServers.github.env.GITHUB_PERSONAL_ACCESS_TOKEN
   • NETLIFY_AUTH_TOKEN → mcpServers.netlify.env.NETLIFY_AUTH_TOKEN
   • PERPLEXITY_API_KEY → mcpServers.perplexity.env.PERPLEXITY_API_KEY
   • FIREFLIES_TOKEN → from mcpServers.fireflies.args, find the string that starts with "Authorization: Bearer " and extract everything after that prefix

2. For Supabase, extract from any Supabase entry in .mcp.json → mcpServers.<any-supabase-server>.env.SUPABASE_ACCESS_TOKEN. This is reliable because it's already a clean JSON string value. If no Supabase server was added, fall back to Vault: $("$OUTWORKOS_ROOT/scripts/get-secret.sh" supabase_access_token).

Important: Prefer reading from .mcp.json JSON values or Vault for all credential access.

Context7 test (MCP-based, not curl):

Context7 has no public REST endpoint, but the MCP server is typically already connected in the current session. Test it by making a lightweight MCP tool call:

1. Use ToolSearch to load mcp__context7__resolve-library-id
2. Call it with libraryName: "nextjs" and query: "test connection"
3. If it returns library results → pass
4. If it errors (auth failure, timeout) → fail — API key invalid, check with Upstash

Run this in parallel with the curl tests above (it's independent).

Services NOT tested (note in report but don't fail):

• Google Workspace: Uses OAuth refresh tokens — the workspace-mcp server handles token refresh internally. If the refresh token is expired, the server will prompt for re-auth on first use.

Handling failures:

For each failed test:

1. Report which service failed and the HTTP status code received
2. Show the specific URL where the user can regenerate the token
3. After listing all failures, ask: "Would you like to update any of these tokens now?"
4. If yes, for each token the user provides:
   • Update the value in Vault via set-secret.sh and/or .mcp.json as appropriate
   • Re-run that specific test to confirm the new token works
   • Repeat until the token passes or the user chooses to skip
5. If all tests pass, continue to Step 8

8. Todoist Sections (if applicable)

If a Todoist project ID is already known for this project (e.g., from a previous /context-map run or from the user), ensure the standard status sections exist:

1. Fetch sections via Sync API: resource_types: ["sections"]
2. Check if "Triage" section exists for the project ID
3. If missing, create all four sections (Triage, Do, Waiting, Defer) in a single batch:

   import json, uuid
   sections = ["Triage", "Do", "Waiting", "Defer"]
   commands = [{"type": "section_add", "uuid": str(uuid.uuid4()), "temp_id": str(uuid.uuid4()), "args": {"name": s, "project_id": PROJECT_ID}} for s in sections]
   with open("/tmp/todoist_batch.json", "w") as f:
       json.dump({"commands": commands}, f)
   

4. If no Todoist project ID is known yet, skip — /context-map will handle section creation when it sets up the Todoist project

Standard sections:

Section	Purpose
Triage	New tasks, needs decision
Do	Committed, actively working
Waiting	Blocked on someone else
Defer	Parked for later

9. GitHub Backup Repository (Optional)

Ask the user: "Do you want to set up a GitHub backup repository for this project?"

If no: skip to Step 10.

If yes:

9a. Prerequisites

1. Verify gh CLI is available: gh --version
2. Verify gh is authenticated: gh auth status
3. If either fails, stop and tell the user to install/authenticate gh

9b. Detect Existing Git State

1. Check if .git/ already exists in the project directory
   • If yes, ask: "This directory already has a git repo. Skip GitHub backup setup?" — if they want to skip, go to Step 10
2. Scan for subdirectories that contain their own .git/ (app repos):

   find . -mindepth 2 -maxdepth 3 -name ".git" -type d
   

3. Record these paths — they'll be excluded via .gitignore

9c. Generate .gitignore

Create .gitignore with:

# Local config
.mcp.json

# OS files
.DS_Store
Thumbs.db

# Dependencies and caches
node_modules/
__pycache__/
.cache/
*.pyc

# Large binary files
*.zip
*.tar.gz
*.dmg

For each subdirectory with its own .git/ detected in 9b, append a line:

# App repo (has its own git)
subdirectory-name/

9d. Initialize and Push

1. git init
2. Derive repo name from the project directory name (lowercase, hyphens for spaces, strip special chars)
3. git add .
4. git commit -m "Initial project backup"
5. gh repo create MattVOLTA/<repo-name> --private --source=. --push
6. Verify: gh repo view MattVOLTA/<repo-name> --json url -q .url

9e. Create Backup Script

Write .claude/hooks/backup.sh:

#!/bin/bash
# Auto-backup project to GitHub on session end
cd "$CLAUDE_PROJECT_DIR" || exit 0

# Skip if no git repo
[ -d .git ] || exit 0

# Skip if no changes
if git diff --quiet HEAD 2>/dev/null && [ -z "$(git status --porcelain)" ]; then
  exit 0
fi

git add -A
git commit -m "Auto-backup $(date +%Y-%m-%d\ %H:%M)" --no-gpg-sign 2>/dev/null
git push origin HEAD 2>/dev/null

Make it executable: chmod +x .claude/hooks/backup.sh

9f. Add SessionEnd Hook

Read the project's .claude/settings.json (create if missing). Add a SessionEnd hook entry:

{
  "hooks": {
    "SessionEnd": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "\"$CLAUDE_PROJECT_DIR/.claude/hooks/backup.sh\"",
            "timeout": 30
          }
        ]
      }
    ]
  }
}

If settings.json already has hooks (e.g., SessionStart), merge the new SessionEnd entry alongside existing hooks — do NOT overwrite.

9g. Verify

Confirm the repo exists and the hook file is in place:

• gh repo view MattVOLTA/<repo-name> --json url
• ls -la .claude/hooks/backup.sh

10. Report

Confirm all created files exist by reading them. Then print a summary:

Project setup complete!

Directory: $OUTWORKOS_PARENT/<project-name>/

Files created:
  .mcp.json
  .claude/commands → $OUTWORKOS_ROOT/skills/  (symlink — all skills available)

Connection tests:
  ✓ GitHub
  ✓ Supabase
  · Google Workspace (OAuth — tested on first use)
  ...

MCP servers configured:
  supabase, google-workspace, github, ...

GitHub backup:
  ✓ Repo: github.com/<user>/<repo-name> (private)
  ✓ SessionEnd hook: .claude/hooks/backup.sh

If GitHub backup was skipped, show:

GitHub backup:
  · Skipped

Use ✓ for passed, ✗ for failed (with note), · for skipped/untestable.

Next step: Suggest: "Run cd $OUTWORKOS_PARENT/<project-name> && claude then /context-map to configure this project."