autonomy

To run autonomous research loops, ask your Claude Code agent to turn on the autonomous loop for the target Synapse project UUID. The skill drives in-session iteration by proposing experiments, dispatching sub-agents, monitoring progress, and updating project synthesis until stop conditions are met.

Sub-agent dispatch and monitoring automatically schedules heartbeat monitoring for long-running experiments and spawns Task-tool sub-agents for pending experiments. This mechanism ensures continuous progress tracking and execution coordination within the autonomous research loop.

No, autonomous research orchestration requires mutual exclusion against server-side autonomous loops. The in-session control flow manages compute availability, experiment status queues, and synthesis updates independently to prevent conflicts with server-side processes.

Project synthesis maintenance updates the platform's project synthesis continuously during experimentation. This ensures the main orchestration loop maintains a rolling, tool-based understanding of observed results and what has been learned throughout the research cycle.

Yes, MCP tool calls are required for project context, experiment proposal, dispatch, and checkpointing. The autonomous research loop relies on in-session control flow and MCP tools to coordinate compute scheduling and stop-condition handling.