Autonomous Agent Instructions

Project: flow

This project is configured for autonomous AI agent workflows with human-in-the-loop approval.

Response Format

Every response MUST end with exactly one of these signals on the final line:

Success signals

done.

Use when task completed successfully with high certainty. No further action needed.

done: <message>

Use when task completed with context to share. Example: done: Added login command with --token flag

Needs human input

needsUpdate: <message>

Use when you need human decision or action. Example: needsUpdate: Should I use OAuth or API key auth?

Error signals

error: <message>

Use when task failed or cannot proceed. Example: error: Build failed - missing dependency xyz

Rules

1. Always end with a signal - The last line must be one of the above
2. One signal only - Never combine signals
3. Be specific - Include actionable context in messages
4. No quotes - Write signals exactly as shown, no wrapping quotes

Examples

Successful implementation

Added the new CLI command with all requested flags.

done.

Completed with context

Refactored the auth module to use the new token format.

done: Auth now supports both JWT and API key methods

Need human decision

Found two approaches for caching:
1. Redis - better for distributed systems
2. In-memory - simpler, faster for single instance

needsUpdate: Which caching approach should I use?

Error occurred

Attempted to run tests but encountered issues.

error: Test suite requires DATABASE_URL environment variable