Resume sessions and handle lost work

Overview

Harness snapshots hold todos, modes and delegation records; chat messages remain in the history provider. In-flight tasks are live process objects and cannot simply be serialized back into execution. Restoring a plan and replaying a side effect are different operations.

A snapshot preserves state, not live tasks

A harness snapshot can retain todos, mode and delegation records across turns. Chat history is retained separately by its provider. A live Task object executing an external request cannot be recreated merely by deserializing a snapshot. After process loss, a delegation can be recorded as lost even while an external system already completed its effect.

Reconcile before replaying an uncertain effect

An interrupted “close request” operation is not equivalent to an operation that was never sent. Inspect its stable operation identifier or authoritative business record before retrying. A recovery response should name lost work and the next decision needed, rather than re-run every unfinished todo automatically. Durable storage is required to test process restart recovery meaningfully.

Resume a saved investigation

1. Run through the wrapper and leave session persistence enabled. Inspect remaining todos before ending the turn.
2. After enabling persistent Orleans storage, restart between turns and verify restored plan/session state. Include DescribeLostDelegations in the user-visible recovery report.
3. Choose an explicit retry/reconciliation path for Lost work. For uncertain external effects, query the recorded operation/result before attempting anything again.

public override async Task<AgentMessage> OnMessage(AgentMessage message)
{
    SetStatusMessage("Planning...");

    var run = await harness.RunAsync(message);
    var text = run.Text;

    // Delegations stranded by a restart — see references/durability.md.
    if (harness.DescribeLostDelegations() is { } lost)
    {
        text += $"{Environment.NewLine}{Environment.NewLine}{lost}";
    }

    var remaining = await harness.GetRemainingTodosAsync();
    if (remaining.Count > 0)
    {
        text += $"{Environment.NewLine}{Environment.NewLine}Not completed within the iteration budget:{Environment.NewLine}"
            + string.Join(Environment.NewLine, remaining.Select(item => $"- {item.Title}"));
    }

    SetStatusMessage(string.Empty);

    var response = message.Response();
    response.Message = text;
    return response;
}

Go deeper

Explore the related documentation.

• FabrCore Harness
• FabrCore Harness Skills
• FabrCore Compaction
• Reporting Honestly