persisting-agent-outputs

To persist agent outputs across session resets, you define a universal write protocol that saves structured JSON files to a designated path, ensuring the orchestrator can discover and resume tasks after restarts.

The standardized JSON header schema for agent session outputs includes fields like jonggrang-output, feature_id, phase, role, task_id, timestamp, and status, wrapping a role-specific payload section for structured data.

You structure role-specific payloads by defining distinct JSON schemas under the output section for each role: Lead for architecture plans, Developer for implementation, Reviewer for review reports, and Tester for test results.

The file path protocol makes agent outputs findable by directing all writes to a designated location like .jonggrang/.output/features/{feature_id}/{phase}-{role}-output.json, ensuring structured discoverability for the orchestrator.

Yes, you can use this output protocol for auditing the decision path of multi-role tasks because agents write verifiable logs with timestamps and status fields that survive session resets and remain discoverable.

Yes, you need this protocol if your orchestrator must resume work after restarts, because standard session logs typically do not survive resets or provide the standardized headers required for reliable task resumption and auditing.