Autonomous Agent Instructions
This project is configured for autonomous AI agent workflows with human-in-the-loop approval.
Response Format
Every response must end with one of these signals:
notify: <message>
Use when you want to tell something to the human. This shows a widget for the user to acknowledge.
Examples:
notify: Task completed successfully
notify: Found 3 potential issues in the code
notify: Need clarification on the database schema
notify: Waiting for API key to be configured
runFlowTask: <task>
Use when the user's original request seems incomplete and a follow-up flow task should run. This shows a widget where the user can approve running the task.
Examples:
runFlowTask: f test
runFlowTask: f build
runFlowTask: f deploy
runFlowTask: f commitWithCheck
Guidelines
- Always end with a signal - Never leave a response without
notify:orrunFlowTask: - Be specific - Messages should clearly describe what happened or what's needed
- Chain tasks - If you think another task should follow, use
runFlowTask: - Report issues - Use
notify:for errors, warnings, or anything requiring human attention
Common Patterns
After completing work
notify: Implemented the new feature as requested
After completing work that needs testing
runFlowTask: f test
After making changes that should be committed
runFlowTask: f commitWithCheck
When encountering an error
notify: Build failed - missing dependency 'foo'
When blocked
notify: Cannot proceed - need database credentials
Available Flow Tasks
Run f tasks to see available tasks for this project.
Common tasks:
f build- Build the projectf test- Run testsf commit- AI-powered commitf commitWithCheck- Commit with Codex code reviewf deploy- Deploy (if configured)