Resume agent coordination after a process failure

Use persistent coordination when an agent process can stop after saving its work but before acknowledging completion. The AgentPlat example recovers that specific boundary with the same operation identity and completed Room runs.

Run the failure scenario

Use the current source checkout, Node.js 22.13+, Corepack-managed pnpm and PostgreSQL 16, or Docker Engine with Compose v2. First complete the build and database setup in persistent collaboration. No real model or external tool service is involved.

From the repository root, with the development PostgreSQL variables set:

corepack pnpm --dir examples/rooms-api demo:recover

For the Docker path, after starting its API container:

docker compose -f examples/rooms-api/compose.yaml exec api pnpm --dir examples/rooms-api demo:recover

What the scenario does

1. Creates a private temporary PostgreSQL schema and starts a child worker.
2. Stops that child after the Room result commits, before coordination acknowledgement.
3. Starts a successor after the coordination lease expires.
4. Checks the retained operation and run identities and asserts two completed runs and two artifacts, reusing completed research and drafting work.
5. Removes the scenario schema during cleanup.

Read the launcher (opens in a new tab) and worker (opens in a new tab) for the failure injection and assertions. A successful exit means those assertions passed. Preserve error output and environment details for a failed run.

What to carry into your application

Persist coordination and results together through their documented adapters. Keep stable work identities across retries. Configure lease ownership and expiry rather than starting a second uncoordinated execution after a timeout. Test your own crash boundaries, including ambiguous external outcomes.

This scenario does not establish recovery of an arbitrary in-flight provider request, exactly-once external effects, availability under all failures or a throughput claim. See integration controls for effect-specific responsibilities and distributed coordination when independent peers and AgentPlat Agent Mesh are required.

The validation record identifies the executed environment. The component maturity matrix separates source behavior from registry distribution and operational evidence.