What problem does it solve?
When multiple AI agent sessions work in parallel, they often collide on the same files or act on wrong assumptions about each other's changes. This Skill delivers verified facts—completion notices, file-touch declarations, and handoffs—to the correct peer session so agents coordinate without guessing.
Core Features & Use Cases
- Recipient Resolution: Lists running sessions via
bin/harness session list and resolves the correct team and agent columns, avoiding silent delivery failures caused by confusing session_id with agent.
- Inbox Messaging: Sends structured messages with
bin/harness inbox send using team, sender, recipient, subject, and body fields, with sender identity pulled from HARNESS_LIVEMSG_TEAM and HARNESS_LIVEMSG_AGENT environment variables.
- Send Discipline: Enforces a policy of sending only confirmed facts (completions, declarations, handoffs) and never speculation or unverified claims, based on CooperBench findings that false assumptions cause most multi-agent failures.
- Use Case: After finishing a refactor of a shared module, notify the other agent session that the file is stable and which areas you will touch next, so it does not edit the same code concurrently.
Quick Start
Ask the agent to notify the other running session that your task is complete and which files you plan to modify next.