harness:session:start

Multi-agent orchestration coordinates complex engineering tasks by managing session lifecycles, assigning file-specific tasks, and delivering signals across Claude, Codex, Gemini, Copilot, and OpenCode agents. A leader plans work, assigns it by file isolation, and monitors progress until completion.

Session-based task assignment is a process where a human leader breaks engineering goals into isolated, file-specific tasks with acceptance criteria and assigns them to appropriate agents. It enforces actor-based mutating operations to prevent conflicts and unsafe local changes during development.

Yes, you can run a coordinated development session across Claude, Codex, and Copilot agents. The leader creates tasks, waits for workers to join, assigns work by file isolation, and monitors progress until completion, ensuring role-based coordination across all agents.

Start a new session with a concise goal, plan discrete tasks, ask the user to spawn worker agents to join, and wait for at least one worker before assigning tasks. The system manages the session lifecycle and ensures actor-authorized mutating operations throughout.

During an orchestration session, you can deliver inject_context, request_action, pause/resume, and abort signals. These signals allow you to observe live activity and control agent behavior while receiving JSON machine-readable outputs for automation.

Preventing agents from editing files directly during multi-agent orchestration avoids conflicts and unsafe local changes. Instead, a human leader coordinates the session, delegates file-specific tasks, and enforces actor-based mutating operations to maintain safe, isolated development workflows.