Shared memory and context tools for agentic work.
Code Rooms
{
"recorded_at": "2026-03-25T00:09:51.483099+00:00",
"scenario_id": "warm_perspective_dead_end_recovery",
"scenario_name": "Warm recovery from exhausted perspective dead end",
"scenario_tags": [
"warm-graph",
"recovery",
"perspective",
"stateful-navigation",
"dead-end"
],
"mode": "manual",
"execution_origin": "replay",
"source_ref": "docs/benchmarks/events/manual-perspective-dead-end-recovery.json",
"cold_graph_time_ms": null,
"warm_graph_time_ms": null,
"time_to_first_good_answer_ms": 5.6,
"time_to_full_proof_ms": 8.2,
"answer_quality": "high",
"plan_changed": false,
"false_start_count": 1,
"tests_identified_before_edit": 0,
"public_claim_worthy": true,
"workflow_notes": "manual recovery reconstructs dead-end unwind by hand before using perspective_back",
"notes": "",
"events": [
"tool_name": "shell.grep",
"query": "search perspective dead-end recovery behavior and check whether back navigation is the safe next move",
"elapsed_ms": 2.4,
"payload_chars": 420,
"surfaced_files": [
"m1nd-mcp/src/perspective_handlers.rs"
"notes": "Manual recovery starts with rediscovering how exhausted branches are meant to unwind.",
"event_index": 1,
"opened_files": []
},
"tool_name": "shell.notes",
"query": "infer that perspective.back is safer than restarting perspective.start after hitting a dead-end branch",
"elapsed_ms": 1.6,
"payload_chars": 260,
"notes": "The operator rebuilds the recovery plan by hand instead of following tool-emitted guidance.",
"event_index": 2,
"opened_files": [],
"surfaced_files": []
"tool_name": "m1nd.perspective_back",
"query": "manually restore previous focus after exhausted branch",
"elapsed_ms": 1.7,
"payload_chars": 340,
"opened_files": [
"notes": "Back navigation works, but only after a manual rediscovery step.",
"event_index": 3,
}
"files_opened": 1,
"repeat_reads": 1,
"search_iterations": 1,
"chars_surfaced": 1020,
"token_proxy": 255,
"guidance_events": 0,
"guidance_followed": 0,
"guidance_followthrough_rate": null,
"reactivated_nodes": 0,
"resume_hints": 0,
"proof_states": [],
"final_proof_state": null,
"proof_state_counts": {},
"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": 0,
"recovery_followed": 0,
"recovery_followthrough_rate": null,
"repo_path": ".",
"question": "Recover from a perspective branch that reaches a dead end, then restore the previous checkpoint and continue without restarting the investigation.",
"expected_strength": "strong"