harness-state-management

Session persistence is maintained by reading and writing to .harness state and learnings files, which track progress and decisions. You load this persisted state at the beginning of a session and save it to enable seamless handoff.

The best way to resume ongoing work without duplication is to load the restored harness state at session start. This automatically recovers your position, decisions, and blockers so the agent can continue exactly where the previous session ended.

You can track milestones and blockers between agent runs by recording decisions and progress into the state.json file. The harness state management mechanism automatically saves these learnings to maintain project continuity and knowledge retention.

Yes, harness state management governs stream-scoped state handling and permission boundaries to maintain project integrity. This ensures that session handoffs and context restoration remain scoped and secure across different workflow streams.

You need to save state for the next session whenever advancing milestones or recording decisions and blockers. This automatic saving between sessions enables seamless handoff and prevents the loss of accumulated agent context and learnings.