Shared memory and context tools for agentic work.
Code Rooms
{
"schema": "m1nd-agent-reliability-round-v0",
"round_id": "round-20260513T003013Z",
"created_at": "2026-05-13T00:30:21.150691+00:00",
"repo": "/Users/kle1nz/m1nd",
"condition": "host-recovery",
"status": "planned",
"lane_count": 7,
"lanes": [
"lane_id": "m1nd-1",
"arm": "m1nd_available"
},
"lane_id": "m1nd-2",
"lane_id": "m1nd-3",
"lane_id": "control-1",
"arm": "no_m1nd"
"lane_id": "control-2",
"lane_id": "control-3",
"lane_id": "judge-1",
"arm": "adjudication"
}
],
"task_battery": [
"task_id": "multi_repo_orientation",
"title": "Multi-repo orientation",
"prompt": "Identify the correct repo, subsystem, and first files to inspect before proposing any action.",
"expected_evidence": [
"correct repo named",
"first relevant subsystem named",
"file or module evidence cited"
"requires_live_proof": false
"task_id": "wrong_workspace_binding",
"title": "Wrong workspace binding",
"prompt": "Diagnose a likely wrong workspace or stale binding before trusting retrieval results.",
"workspace mismatch named",
"shortest honest recovery route named",
"no false graph-health claim"
"requires_live_proof": true
"task_id": "transport_closed_recovery",
"title": "Transport closed recovery",
"prompt": "Recover or reroute after a dead MCP transport without fabricating success.",
"transport failure detected",
"fallback route recorded",
"missing proof preserved"
"task_id": "stale_runtime_route",
"title": "Stale runtime or PATH route",
"prompt": "Identify stale runtime, PATH shadowing, or tool-surface mismatch and name the repair command.",
"runtime versions compared",
"PATH or configured binary route checked",
"host rebind caveat preserved"
"task_id": "structural_edit_prep",
"title": "Structural edit preparation",
"prompt": "Gather enough connected context to name the safe edit target and focused proof steps.",
"edit target named",
"blast-radius or dependency reasoning cited",
"proof gates named"
"task_id": "root_cause_triage",
"title": "Root-cause triage",
"prompt": "From a realistic symptom, isolate the most likely fault boundary without broad file dumping.",
"suspect boundary named",
"alternative theory rejected or left open",
"next verification command or file named"
"task_id": "continuity_resume",
"title": "Continuity resume",
"prompt": "Continue a partially completed investigation without restarting from zero.",
"prior state restored",
"next unresolved question named",
"unneeded rediscovery avoided"
"success_criteria": [
"lane reaches the correct target or honest recovery route",
"lane does not claim proof it did not gather",
"lane preserves missing proof when proof is missing",
"lane avoids unnecessary restart behavior after useful hints"
"non_claims": [
"no public performance claim is made from one benchmark round",
"agent testimony is not evidence by itself without scored task results",
"m1nd does not replace tests, compiler output, git history, rg, or direct file truth",
"warm-graph results do not equal cold-start behavior",
"host, transport, runtime, and workspace failures must be recorded instead of smoothed away"
]