Shared memory and context tools for agentic work.
Code Rooms
{
"recorded_at": "2026-03-25T00:11:36.428881+00:00",
"scenario_id": "warm_seek_broad_query_recovery",
"scenario_name": "Warm recovery from a broad seek query into exact helper lookup",
"scenario_tags": [
"warm-graph",
"recovery",
"seek",
"intent",
"exact-lookup"
],
"mode": "m1nd_warm",
"execution_origin": "replay",
"source_ref": "docs/benchmarks/events/warm-seek-broad-query-recovery.json",
"cold_graph_time_ms": null,
"warm_graph_time_ms": null,
"time_to_first_good_answer_ms": 1.4,
"time_to_full_proof_ms": 2.8,
"answer_quality": "high",
"plan_changed": false,
"false_start_count": 0,
"tests_identified_before_edit": 0,
"public_claim_worthy": true,
"workflow_notes": "Seek reaches the right subsystem, then the recovery loop narrows into exact helper search before view confirms the proof-state contract.",
"notes": "",
"events": [
"tool_name": "m1nd.seek",
"query": "Where does seek recover when a broad natural-language query needs the exact proof-state fallback helper?",
"elapsed_ms": 0.9,
"payload_chars": 320,
"surfaced_files": [
"m1nd-mcp/src/layer_handlers.rs"
"proof_state": "triaging",
"next_suggested_tool": "view",
"next_suggested_target": "m1nd-mcp/src/layer_handlers.rs",
"next_step_hint": "Open the seek helper logic next to confirm how broad semantic matches turn into a proof-state decision.",
"next_tool_used": "view",
"event_index": 1,
"opened_files": []
},
"tool_name": "shell.notes",
"query": "The semantic pass reached the right subsystem, but the agent still needs the exact helper name to recover cleanly.",
"elapsed_ms": 0.6,
"payload_chars": 220,
"hint": "Recover by exact search on l2_seek_proof_state and l2_seek_next_step before reading the output contract.",
"suggested_next_step": "Retry with an exact helper search instead of another broad semantic probe.",
"usage_hint": "When seek lands in the right neighborhood, use search to pin the exact helper name and avoid file guessing.",
"next_suggested_tool": "search",
"next_tool_used": "search",
"recovery_followed": true,
"event_index": 2,
"opened_files": [],
"surfaced_files": []
"tool_name": "m1nd.search",
"query": "l2_seek_proof_state",
"elapsed_ms": 0.8,
"payload_chars": 470,
"opened_files": [
"proof_state": "proving",
"next_step_hint": "Open the helper definition next to confirm how seek grades blocked, triaging, proving, and ready_to_edit.",
"event_index": 3
"tool_name": "m1nd.view",
"target": "m1nd-mcp/src/layer_handlers.rs",
"elapsed_ms": 0.5,
"payload_chars": 820,
"event_index": 4,
}
"files_opened": 1,
"repeat_reads": 3,
"search_iterations": 2,
"chars_surfaced": 1830,
"token_proxy": 458,
"guidance_events": 3,
"guidance_followed": 3,
"guidance_followthrough_rate": 1.0,
"reactivated_nodes": 0,
"resume_hints": 0,
"proof_states": [
"triaging",
"proving"
"final_proof_state": "proving",
"proof_state_counts": {
"proving": 1,
"triaging": 1
"progress_events": 0,
"max_progress_pct": 0.0,
"active_phases": [],
"next_phases": [],
"progress_event_types": [],
"progress_delivery_modes": [],
"live_progress_events": 0,
"replay_progress_events": 0,
"snapshot_progress_events": 0,
"progress_guidance_events": 0,
"progress_guidance_followed": 0,
"progress_guidance_followthrough_rate": null,
"recovery_events": 1,
"recovery_followed": 1,
"recovery_followthrough_rate": 1.0,
"repo_path": ".",
"question": "Recover from a broad semantic seek that lands in the right subsystem but not the exact helper name, then narrow into `l2_seek_proof_state` and confirm the output contract instead of guessing files.",
"expected_strength": "strong"