What problem does it solve?
Prevents conversational agents from promising actions they do not perform by detecting commitment language in outgoing replies and forcing immediate execution, providing accountability and reducing missed follow-ups.
Core Features & Use Cases
- Immediate Execution: Scan replies for trigger phrases and execute the promised action before sending the message.
- Intent Logging & Recovery: Record pending commitments to a durable commitments.jsonl ledger and recover incomplete actions on startup or heartbeat.
- Safe Failure Handling: If execution is impossible, rewrite the reply to avoid promises and notify the owner with the blocking reason.
- Use Case: When an agent intends to "I'll send the report", it will attach or send the report immediately, log the action, and then respond "Sent ✅" instead of making an unverifiable promise.
Quick Start
Before sending any reply that includes a promise, execute the promised action immediately, append a pending intent to the commitments.jsonl file, and then report the result in your response.