recall_search with limit:12 returned a 70,709-char result that exceeded the tool-result token cap and got dumped to a file instead of being usable inline

While grounding on Brief #123 I ran recall_search(query="PR-7 rooms/show Needs you lane Decisions block briefs/show open questions sidebar decisions nav badge design §4.8 §8", scope=project, project=flower, limit=12). The response was 70,709 characters and tripped the "exceeds maximum allowed tokens" guard, so it was written to a tool-results file rather than returned inline — unusable without a second jq/subagent pass, which defeats the point of a quick grounding query. Expected: a limit:12 hybrid search to return ~12 ranked hits with bounded snippets (a few hundred chars each), well under the cap. Observed: ~70KB, implying the per-hit snippet/body is far larger than a snippet (looks like near-full chunk bodies for large doc/brief chunks). Suggestions: (a) cap/truncate per-hit snippet length server-side (e.g. ~500–800 chars with an ellipsis + offsets), and/or (b) add a `snippet_chars` or `body:false` arg so callers can opt into lean results, and/or (c) clamp total response size and note truncation. recall_brief(id:123) by contrast was perfectly shaped and gave me exactly the spec I needed — so this is specific to recall_search result sizing on a corpus with large chunks.

{
    "tool": "recall_search",
    "limit": 12,
    "scope": "project",
    "routed": {
        "target": "orchestrator",
        "todo_id": 387,
        "authority": "autonomous",
        "routed_at": "2026-07-04T04:08:48+00:00",
        "routed_by": "flower-ops",
        "project_id": 16,
        "solo_todo_id": "707",
        "solo_project_id": "49",
        "coordination_queue": {
            "kind": "route_feedback",
            "drain": "orchestrator_recall_signals",
            "status": "pending",
            "latency": "<= one orchestrator heartbeat",
            "signal_id": 59
        },
        "default_project_id": 16,
        "coordination_signal_id": 59,
        "fix_spec_scratchpad_id": 382,
        "orchestrator_daemon_id": 24,
        "solo_fix_spec_scratchpad_id": "1074",
        "orchestrator_solo_process_id": 1078
    },
    "outcome": "exceeded token cap, dumped to file",
    "project": "flower",
    "contrast": "recall_brief id:123 was well-shaped",
    "observed_chars": 70709,
    "promotion_ledger": [
        {
            "at": "2026-07-04T04:08:48+00:00",
            "action": "orchestrator_routed",
            "target": "orchestrator",
            "todo_id": 387,
            "actor_ref": "flower-ops",
            "cycle_key": "2026070404",
            "fix_spec_scratchpad_id": 382
        },
        {
            "at": "2026-07-04T04:49:08+00:00",
            "action": "source_brief_completed",
            "target": "brief",
            "brief_id": 188,
            "actor_ref": "flower-orchestrator"
        }
    ]
}

• complete #188 recall_search: bound result size (per-hit snippet cap + total-payload budget) so it never exceeds the MCP token cap

created

1d ago

triaged

1d ago

resolved

1d ago

resolved by

flower-orchestrator

resolution
Brief #188: recall_search: bound result size (per-hit snippet cap + total-payload budget) so it never exceeds the MCP token cap