Skip to main content

Friday Agent

Friday Agent is the flagship AI agent in Friday Dev, powered by advanced AI with custom tooling for autonomous development.

Overview​

Friday Agent combines:

  • Advanced AI - Best-in-class reasoning
  • Custom tools - File operations, git, terminal
  • Autonomous execution - Minimal human intervention
  • Safety features - Controlled access and rollback

Capabilities​

What Friday Agent Can Do​

  • ✅ Read and understand entire codebases
  • ✅ Plan multi-file implementations
  • ✅ Write and modify code
  • ✅ Run tests and fix failures
  • ✅ Create pull requests
  • ✅ Respond to code review feedback
  • ✅ Install dependencies
  • ✅ Run build commands

Best Use Cases​

Use CaseEffectiveness
New features⭐⭐⭐⭐⭐
Bug fixes⭐⭐⭐⭐⭐
Refactoring⭐⭐⭐⭐
Documentation⭐⭐⭐⭐
Testing⭐⭐⭐⭐
DevOps⭐⭐⭐

Setup​

API Key​

Friday Agent requires an Anthropic API key:

export ANTHROPIC_API_KEY="your-anthropic-api-key"

Get your key at console.anthropic.com.

Configuration​

{
  "agents": {
    "friday": {
      "enabled": true,
      "model": "claude-sonnet-4-20250514",
      "autonomy": "workspace-write",
      "maxTokens": 8192
    }
  }
}

Usage​

From UI​

  1. Open a task
  2. Click "Run Agent"
  3. Select "Friday Agent"
  4. Click "Start"

From CLI​

# Run on a task
friday-dev run --task 123 --agent friday

# Run with specific autonomy
friday-dev run --task 123 --agent friday --autonomy workspace-write

How It Works​

1. Task Understanding​

┌─────────────────────────────────────────────────────────────────┐
│ Friday Agent reads:                                              │
│                                                                  │
│ • Task title and description                                     │
│ • Codebase structure (file tree)                                │
│ • Relevant source files                                          │
│ • Package.json / Cargo.toml / etc.                              │
│ • Existing tests                                                 │
│ • Git history (recent commits)                                   │
└─────────────────────────────────────────────────────────────────┘

2. Planning​

Friday Agent creates a step-by-step plan:

Plan for: Add user authentication
──────────────────────────────────
1. Create User model in src/models/user.ts
2. Add auth routes in src/routes/auth.ts
3. Implement JWT middleware in src/middleware/auth.ts
4. Update database schema
5. Add tests in tests/auth.test.ts
6. Update README with auth documentation

3. Execution​

The agent executes each step, showing progress:

[1/6] Creating User model...
      ✓ Created src/models/user.ts (+45 lines)

[2/6] Adding auth routes...
      ✓ Created src/routes/auth.ts (+78 lines)

[3/6] Implementing JWT middleware...
      ✓ Created src/middleware/auth.ts (+32 lines)

[4/6] Updating database...
      ✓ Modified prisma/schema.prisma (+12 lines)
      ✓ Running prisma migrate dev...

[5/6] Adding tests...
      ✓ Created tests/auth.test.ts (+120 lines)
      ✓ Running tests... 12 passed

[6/6] Updating documentation...
      ✓ Modified README.md (+25 lines)

✓ Task complete! Created PR #456

4. Output​

After completion:

  • Diff view of all changes
  • Logs of all actions
  • PR link (if configured)
  • Test results (if tests exist)

Autonomy Levels​

{
  "autonomy": "workspace-write"
}
  • ✅ Read any file
  • ✅ Write to workspace files
  • ✅ Run build/test commands
  • ❌ Install global packages
  • ❌ Modify system files
  • ❌ Access network (except git)

skip-permissions-unsafe​

{
  "autonomy": "skip-permissions-unsafe"
}
  • ✅ Full file system access
  • ✅ Run any command
  • ✅ Install packages
  • ⚠️ Use with caution in trusted environments only

Advanced Features​

Iterative Development​

Friday Agent can iterate based on feedback:

  1. Agent completes task
  2. You review and comment
  3. Click "Re-run with feedback"
  4. Agent addresses comments

Multi-file Understanding​

Friday Agent reads multiple files to understand context:

Reading context:
├── src/index.ts (entry point)
├── src/config.ts (configuration)
├── src/types.ts (type definitions)
├── package.json (dependencies)
└── tsconfig.json (TypeScript config)

Test-Driven Development​

If tests exist, Friday Agent:

  1. Reads existing tests
  2. Runs tests before changes
  3. Makes changes
  4. Runs tests after changes
  5. Fixes any failures

Tips for Best Results​

1. Write Clear Tasks​

Good:

## Task
Add password reset functionality

## Requirements
- User requests reset via email
- Send email with reset link (valid 24h)
- Reset page validates token and updates password
- Use existing email service (src/services/email.ts)

## Files to reference
- src/routes/auth.ts (existing auth routes)
- src/models/user.ts (user model)

Bad:

Add password reset

2. Provide Context​

Mention:

  • Related files
  • Existing patterns to follow
  • Constraints or requirements
  • Technology preferences

3. Break Down Large Tasks​

Instead of:

Build complete user management system

Create subtasks:

1. Add user registration
2. Add user login
3. Add password reset
4. Add profile management
5. Add admin panel

Troubleshooting​

Agent Stops Unexpectedly​

  1. Check API key balance
  2. Look for error in logs
  3. Try running again
  4. Contact support if persists

Output Quality Issues​

  1. Add more detail to task
  2. Include example code
  3. Reference similar files
  4. Try breaking into subtasks

Slow Performance​

  1. Reduce codebase size (use .gitignore)
  2. Be specific about files to modify
  3. Use lighter agent for simple tasks

Pricing​

Friday Agent uses Claude API:

Next Steps​