Shared memory and context tools for agentic work.
Code Rooms
{
"schema": "m1nd-real-world-agent-round-v0",
"round_id": "real-world-20260513T005733Z",
"created_at": "2026-05-13T01:04:21.132054+00:00",
"status": "planned",
"fixture_repos": [
"repo_id": "click-python-cli",
"url": "https://github.com/pallets/click.git",
"ecosystem": "python",
"why": "mature Python CLI library with docs, tests, commands, options, and parser internals",
"local_path": ".m1nd-benchmark-fixtures/real-world/click-python-cli"
},
"repo_id": "p-limit-node",
"url": "https://github.com/sindresorhus/p-limit.git",
"ecosystem": "typescript",
"why": "small TypeScript utility with compact source, tests, package metadata, and public API docs",
"local_path": ".m1nd-benchmark-fixtures/real-world/p-limit-node"
"repo_id": "human-panic-rust-cli",
"url": "https://github.com/rust-cli/human-panic.git",
"ecosystem": "rust",
"why": "Rust library/CLI-adjacent crate with examples, feature flags, tests, and public docs",
"local_path": ".m1nd-benchmark-fixtures/real-world/human-panic-rust-cli"
}
],
"lane_workspace_root": ".m1nd-benchmark-fixtures/real-world-lanes/real-world-20260513T005733Z",
"task_count": 10,
"task_matrix": [
"task": {
"task_id": "repo_architecture_audit",
"mode": "audit",
"title": "Architecture audit",
"prompt": "Explain the repo architecture, main modules, entrypoints, data/control flow, and top risks.",
"expected_evidence": [
"main entrypoints named",
"module boundaries named",
"at least two real file references",
"risk list separates proven facts from hypotheses"
"requires_code_change": false
"repo": {
"task_id": "feature_location",
"mode": "localize",
"title": "Feature localization",
"prompt": "Find where a named feature or public behavior is implemented and identify the tests that protect it.",
"implementation file named",
"test file named or missing test stated",
"false-positive files avoided"
"task_id": "flow_explanation",
"mode": "explain",
"title": "End-to-end flow explanation",
"prompt": "Explain a realistic request/command/API flow from public entrypoint to internal behavior.",
"entrypoint named",
"intermediate calls named",
"observable output or side effect named"
"task_id": "bug_symptom_triage",
"mode": "diagnose",
"title": "Bug symptom triage",
"prompt": "Given a realistic symptom, isolate the most likely fault boundary and name the next verification step.",
"most likely fault file or function named",
"alternative theory preserved or rejected",
"next command/test/file named"
"task_id": "safe_change_plan",
"mode": "plan",
"title": "Safe change plan",
"prompt": "Plan a small behavior change, including blast radius, files to edit, and proof gates.",
"edit targets named",
"downstream callers or tests named",
"risky assumptions explicit"
"task_id": "small_feature_patch",
"mode": "patch",
"title": "Small feature patch",
"prompt": "Implement a tiny feature or option consistent with local style and run focused checks.",
"minimal patch",
"test or example updated when appropriate",
"focused check result recorded"
"requires_code_change": true
"task_id": "seeded_bug_fix",
"title": "Seeded bug fix",
"prompt": "Fix a seeded or clearly described bug without broad refactors.",
"root cause named",
"patch is scoped",
"regression proof recorded"
"task_id": "bounded_refactor_plan",
"title": "Bounded refactor plan",
"prompt": "Prepare a bounded refactor and identify hidden coupling before any edit.",
"coupled files named",
"safe ordering proposed",
"rollback or proof boundary named"
"task_id": "code_review_diff",
"mode": "review",
"title": "Code review of a diff",
"prompt": "Review a supplied or seeded diff for real bugs, regressions, and missing tests.",
"findings ordered by severity",
"file/line references when available",
"style-only comments avoided"
"task_id": "docs_drift_check",
"mode": "docs",
"title": "Docs/spec drift check",
"prompt": "Compare README/docs claims against implementation and identify drift or missing documentation.",
"claim source named",
"code truth named",
"drift or no-drift conclusion justified"
"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"
"success_criteria": [
"correct files or modules identified with evidence",
"causal explanation matches actual code paths",
"missing tests and missing proof are named honestly",
"patch tasks stay minimal and are verified with focused checks",
"control and m1nd lanes use comparable repo snapshots"
"non_claims": [
"one real-world round is not a public performance claim",
"benchmark repos are fixtures, not proof of universal repo performance",
"m1nd does not replace tests, compiler output, git history, rg, or direct file truth",
"agent testimony is not evidence without scored task artifacts",
"warm-graph and cold-graph results must be reported separately",
"a correct plan is not the same evidence as a correct patch"
]