Ambient Swarm Autonomous Work-Claim Daemon & Real-Time Event Loop Engine

Overview

The Ambient Swarm Work-Claim Daemon provides continuous autonomous work-claim and dynamic seat dispatch within the ZQK Knowledge Kernel. Operating inside the host AmbientEngine supervisor tick loop, the daemon coordinates with ConsensusArbiter to ensure safe, race-free, and exclusive task leasing across registered agent peer seats.

Architecture

flowchart TD
    subgraph Engine ["Ambient Engine"]
        ST["Supervise Tick Loop"]
        BO["BacklogObserver / SwarmDispatcher"]
        SL["SwarmLoopState (Unassigned Queue & Retries)"]
        CA["ConsensusArbiter (Atomic Task Leases)"]
        ES["Real-time Event Stream (TaskEvents)"]
    end

    subgraph Seats ["Agent Swarm Peer Seats"]
        S1["Peer Seat: Software Engineer"]
        S2["Peer Seat: QA Sentinel"]
        S3["Peer Seat: TPM Coordinator"]
    end

    ST -->|superviseTick| BO
    BO -->|AutoClaimTick| SL
    SL -->|NextUnassigned| BO
    BO -->|LeaseTask| CA
    BO -->|PublishEvent| ES
    BO -->|DispatchLease| Seats
    Seats -->|CompleteTask / YieldTask| BO

Task Lifecycle & State Transitions

  1. Unassigned / Enqueued: Tasks are enqueued via EnqueueTask(taskID). If not already leased or quarantined, the task enters SwarmLoopState.unassigned.

  2. Auto-Claim Dispatch: During each supervisor tick (superviseTick / AutoClaimTick), the dispatcher iterates healthy peer seats (HealthySeats()). If a seat's active lease count is below MaxClaimsPerSeat, a lease is atomically requested via ConsensusArbiter.LeaseTask. Upon acquisition, a TaskEventClaimed event is broadcast to the TaskEvents stream.

  3. Task Completion: When a peer seat finishes work, it invokes CompleteTask(taskID, seatID). The dispatcher verifies seat lease ownership, commits the task in ConsensusArbiter.CommitTask, evicts the local active lease, and broadcasts TaskEventCompleted.

  4. Task Yield & Escalation: If a peer seat encounters an unresolvable obstacle or boundary exception, it calls YieldTask(taskID, seatID, reason). The lease is revoked, and the task's retry count is incremented:

  • If retryCount <= MaxRetries, the task is re-queued into unassigned for redistribution, emitting TaskEventYielded.
  • If retryCount > MaxRetries, the task transitions to quarantine status, emitting TaskEventQuarantined to prevent infinite thrashing loops.

Telemetry & Verification

All lifecycle events are streamed non-blockingly over <-chan SwarmTaskEvent with payload metadata including TaskID, SeatID, EventType, RetryCount, Reason, and Timestamp.