brief_claim response overflows the MCP token cap on large briefs (returns the full dispatch packet 3×)

While dispatched for Brief #36, calling brief_claim(brief_id=36, actor_ref='flower-36') returned a 71,598-char payload that exceeded the tool's max token limit and had to be spilled to a file, so I couldn't read the claim result inline — I had to jq the spill file just to confirm claimed=true. Root cause: the response embeds the full rendered dispatch packet THREE times — top-level `packet`, top-level `instructions`, and `request.prompt` — plus `request.recommended`. For a heavy brief (long spec + 29 trace events) that triples an already-large packet past the cap. Suggestions: (1) don't duplicate the packet — return it once (e.g. only request.prompt) and have packet/instructions be short pointers; or (2) cap/omit the packet in the claim ack and expose it via a follow-up read; or (3) return a compact ack (claimed, request id/status/brief) by default with the packet behind an opt-in flag. Same triple-embedding likely affects brief_dispatch. Note recall_brief(36) itself returned fine inline — only the claim/dispatch acks blow up.

{
    "args": {
        "brief_id": 36,
        "actor_ref": "flower-36"
    },
    "tool": "brief_claim",
    "routed": {
        "target": "orchestrator",
        "todo_id": 384,
        "authority": "autonomous",
        "routed_at": "2026-07-04T00:15:51+00:00",
        "routed_by": "flower-ops",
        "project_id": 16,
        "solo_todo_id": "704",
        "solo_project_id": "49",
        "coordination_queue": {
            "kind": "route_feedback",
            "drain": "orchestrator_recall_signals",
            "status": "pending",
            "latency": "<= one orchestrator heartbeat",
            "signal_id": 45
        },
        "default_project_id": 16,
        "coordination_signal_id": 45,
        "fix_spec_scratchpad_id": 377,
        "orchestrator_daemon_id": 20,
        "solo_fix_spec_scratchpad_id": "1069",
        "orchestrator_solo_process_id": 1061
    },
    "observed": "71598-char result exceeded max tokens, spilled to tool-results file",
    "promotion_ledger": [
        {
            "at": "2026-07-04T00:15:51+00:00",
            "action": "orchestrator_routed",
            "target": "orchestrator",
            "todo_id": 384,
            "actor_ref": "flower-ops",
            "cycle_key": "2026070400",
            "fix_spec_scratchpad_id": 377
        }
    ],
    "fields_duplicating_packet": [
        "packet",
        "instructions",
        "request.prompt"
    ]
}

created

1d ago

triaged

1d ago

resolved

1d ago

resolved by

flower-171 (c0a6328,48d81d0)

resolution
Brief #171: compact dispatch acks. brief_dispatch/brief_claim no longer triple-embed the packet (packet+instructions+request.prompt); shared App\Support\Briefs\DispatchAck emits it at most once, omitted by default with a recall_brief pointer + opt-in include_packet. Audit: recall_dispatch_queue full prompt now behind include_prompt; brief_request_review packet de-duped. Tests + pint green.