The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.
TypeScript 3d-modelingagent-frameworkagent-harnessai-agentscadclaude-codecodexcoding-agentsdeveloper-toolsesp32game-developmentmcp-servermulti-agentopen-hardwareopencodeopenharnesspcbroboticsterminaltmux
{
"schema_version": 1,
"status": "Synthetic design examples, not production memories or an implemented wire schema.",
"data_policy": "Invented users, projects, source events, revisions, and outcomes. No profile describes the current user. Source roles are fixture data.",
"source_events": [
{
"id": "source_style_a",
"profileId": "example_a",
"role": "user",
"content": "For concurrency bugs, first spell out the state transitions and what must remain true, then reproduce the bug and verify the fix. This helps me review the race."
},
{
"id": "source_style_b",
"profileId": "example_b",
"role": "user",
"content": "For reproducible bugs, start with the smallest failing test, make a small fix, verify it, then explain why it works. Concrete feedback helps me review it."
},
{
"id": "source_cache_decision",
"profileId": "example_service",
"role": "user",
"content": "For this single-process preview server, choose the in-process cache to avoid operating another service. We considered Redis but do not need it yet. We expect fewer repeated renders. Revisit the choice if we add independent workers."
},
{
"id": "source_cache_test",
"profileId": "example_service",
"role": "tool",
"content": "At revision example_r1, in the single-process test fixture: two sequential requests for the same unchanged input invoke the renderer once. Test passed. No latency benchmark or concurrent-worker test was run."
},
{
"id": "source_benchmark_hypothesis",
"profileId": "example_measurement",
"role": "assistant",
"content": "The cache size might be causing the slow queries. I propose testing a larger cache. This is a hypothesis."
},
{
"id": "source_benchmark_result",
"profileId": "example_measurement",
"role": "tool",
"content": "Synthetic benchmark report: revision example_search_r1, workload fixture Q, 100 requests. Run A used cache 128 MB on machine A and median latency was 80 ms. Run B used cache 512 MB on machine B and median latency was 40 ms. Hardware and cache size both changed. No controlled rerun has occurred."
},
{
"id": "source_guide_authority",
"profileId": "example_service",
"role": "user",
"content": "For preview verification, docs/preview-checks.md is our canonical project guide. Read it when changing the preview renderer. Keep setup commands there rather than making another copy."
},
{
"id": "source_guide_observation",
"profileId": "example_service",
"role": "tool",
"content": "In project preview_example, docs/preview-checks.md exists at example_guide_r2. This observation verifies file presence and revision only; host discovery, loading, and command execution have not been checked."
},
{
"id": "source_experiment_objective",
"profileId": "example_measurement",
"role": "user",
"content": "Compare training loss with evaluator eval_v1 on data_v1, lower is better, on GPU_A under a 300-second training budget excluding startup and compilation. This comparison is exploratory, not an adopted architecture decision."
},
{
"id": "source_typed_experiment_runs",
"profileId": "example_measurement",
"role": "tool",
"content": "Synthetic report report_runs_v1: baseline commit train_r1 completed on GPU_A with evaluator eval_v1, data_v1, seed 7, budget 300 training seconds excluding startup and compilation, score 1.2. Candidate train_r2 under the same configured conditions crashed with OOM before evaluation; its CSV score is placeholder 0.0 and no valid score exists. No repeated runs or uncertainty estimates."
}
],
"records": [
{
"schemaVersion": 1,
"id": "memory_style_a",
"revision": 1,
"kind": "working_preference",
"facet": "reasoning_and_feedback",
"assertionType": "stated_preference",
"scope": {
"profileId": "example_a"
},
"claim": "For concurrency bugs, the developer prefers examining the state model before the reproducer and fix.",
"rationale": "The developer says this helps them review the race.",
"futureAction": "Start with the relevant state transitions and invariant, then reproduce and verify the repair.",
"applicability": {
"taskType": "debugging",
"problemClass": "concurrency"
},
"exceptions": [],
"details": {
"style": {
"preferredSequence": [
"state_model",
"reproducer",
"fix",
"verification"
]
}
},
"evidenceClass": "user_stated",
"evidence": [
{
"sourceEventId": "source_style_a",
"relation": "supports",
"paths": [
"/claim",
"/rationale",
"/applicability",
"/details/style"
]
}
],
"state": "active",
"conflictKey": "example_a:concurrency:feedback_sequence",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null
},
"supersedes": []
},
{
"schemaVersion": 1,
"id": "memory_style_b",
"revision": 1,
"kind": "working_preference",
"facet": "reasoning_and_feedback",
"assertionType": "stated_preference",
"scope": {
"profileId": "example_b"
},
"claim": "For reproducible bugs, the developer prefers a minimal failing test before a small fix and explanation.",
"rationale": "The developer says concrete feedback helps them review the work.",
"futureAction": "Begin with the focused reproducer, make the bounded change, and explain its correctness.",
"applicability": {
"taskType": "debugging",
"reproducible": true
},
"exceptions": [],
"details": {
"style": {
"preferredSequence": [
"reproducer",
"bounded_fix",
"verification_and_explanation"
]
}
},
"evidenceClass": "user_stated",
"evidence": [
{
"sourceEventId": "source_style_b",
"relation": "supports",
"paths": [
"/claim",
"/rationale",
"/applicability",
"/details/style"
]
}
],
"state": "active",
"conflictKey": "example_b:reproducible_bug:feedback_sequence",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null
},
"supersedes": []
},
{
"schemaVersion": 1,
"id": "memory_cache_decision",
"revision": 1,
"kind": "project_decision",
"facet": "architecture",
"assertionType": "accepted_decision",
"scope": {
"profileId": "example_service",
"projectId": "preview_example"
},
"claim": "The preview server selected an in-process cache for its single-process deployment.",
"rationale": "Avoid operating another service while the deployment does not need shared cache coordination.",
"futureAction": "Preserve the rationale while these conditions apply; reconsider cache coordination for independent workers.",
"applicability": {
"component": "preview_server",
"deployment": "single_process"
},
"exceptions": [],
"details": {
"decision": {
"problem": "Avoid repeated rendering without adding unnecessary operational work.",
"alternatives": [
{
"option": "in_process_cache",
"disposition": "chosen"
},
{
"option": "Redis",
"disposition": "deferred",
"reason": "Shared coordination is not needed for the current deployment."
}
],
"expectedOutcome": "Fewer repeated renders.",
"observedOutcome": "The sequential unchanged-input fixture invoked the renderer once for two requests.",
"revisitWhen": [
"The deployment gains independent workers."
],
"unknowns": [
"Concurrent-worker behavior",
"End-to-end latency effect"
]
}
},
"evidenceClass": "user_stated",
"evidence": [
{
"sourceEventId": "source_cache_decision",
"relation": "supports",
"paths": [
"/claim",
"/rationale",
"/applicability",
"/details/decision/problem",
"/details/decision/alternatives",
"/details/decision/expectedOutcome",
"/details/decision/revisitWhen"
]
},
{
"sourceEventId": "source_cache_test",
"relation": "supports",
"paths": [
"/details/decision/observedOutcome",
"/details/decision/unknowns"
],
"verification": {
"method": "test",
"result": "passed",
"artifact": "source_cache_test",
"revision": "example_r1",
"environment": "single-process test fixture",
"coverage": "Two sequential requests with the same unchanged input.",
"limitations": [
"No latency benchmark",
"No independent-worker test"
]
}
}
],
"state": "active",
"conflictKey": "example_service:preview_example:cache_choice",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null,
"recheckWhen": [
"Worker topology changes",
"Input invalidation contract changes"
]
},
"supersedes": []
},
{
"schemaVersion": 1,
"id": "memory_open_experiment",
"revision": 1,
"kind": "working_continuity",
"facet": "verification",
"assertionType": "temporary_state",
"scope": {
"profileId": "example_measurement",
"projectId": "search_example",
"taskId": "investigation_example"
},
"claim": "The cache investigation has a comparison that changed both cache size and hardware; the isolated cache effect remains unknown.",
"rationale": "The observed comparison cannot isolate which change caused the difference.",
"futureAction": "Check the current benchmark setup and, if still relevant, compare cache sizes with hardware and workload held constant.",
"applicability": {
"taskId": "investigation_example",
"component": "search"
},
"exceptions": [],
"details": {
"experiment": {
"hypothesis": "A larger cache reduces query latency.",
"hypothesisStatus": "unresolved",
"intervention": [
"Cache size 128 MB to 512 MB",
"Machine A to machine B"
],
"result": "Median latency changed from 80 ms to 40 ms in the reported runs.",
"limitations": [
"Two factors changed",
"No controlled rerun",
"No statistical uncertainty estimate supplied"
]
}
},
"evidenceClass": "observed_verified",
"evidence": [
{
"sourceEventId": "source_benchmark_hypothesis",
"relation": "references",
"paths": [
"/details/experiment/hypothesis"
]
},
{
"sourceEventId": "source_benchmark_result",
"relation": "supports",
"paths": [
"/claim",
"/details/experiment/hypothesisStatus",
"/details/experiment/intervention",
"/details/experiment/result",
"/details/experiment/limitations"
],
"verification": {
"method": "benchmark",
"result": "comparison_observed_cause_unresolved",
"artifact": "source_benchmark_result",
"revision": "example_search_r1",
"environment": "Machines A and B; fixture Q; 100 requests per run",
"coverage": "Reported medians for these two runs only",
"limitations": [
"Hardware and cache size changed together"
]
}
}
],
"state": "active",
"conflictKey": "example_measurement:investigation_example:current_state",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null,
"recheckWhen": [
"Task completes",
"A controlled rerun is available"
]
},
"supersedes": []
},
{
"schemaVersion": 1,
"id": "memory_preview_guide",
"revision": 1,
"kind": "reference",
"facet": "verification",
"assertionType": "project_constraint",
"scope": {
"profileId": "example_service",
"projectId": "preview_example"
},
"claim": "The project identifies docs/preview-checks.md as the canonical preview verification guide.",
"rationale": "The developer wants setup commands maintained in one project location.",
"futureAction": "Read the current guide for preview renderer changes and check actual host delivery before claiming it was loaded.",
"applicability": {
"component": "preview_renderer"
},
"exceptions": [],
"details": {
"reference": {
"resourceType": "canonical_project_guide",
"locator": "docs/preview-checks.md",
"authority": "user_identified_project_guidance",
"sourceRevision": "example_guide_r2",
"readWhen": [
"changing_preview_renderer"
],
"availability": {
"presence": "verified",
"hostDiscovery": "unknown",
"loaded": "unknown",
"execution": "unknown"
}
}
},
"evidenceClass": "user_stated",
"evidence": [
{
"sourceEventId": "source_guide_authority",
"relation": "supports",
"paths": [
"/claim",
"/rationale",
"/applicability",
"/details/reference/authority",
"/details/reference/readWhen"
]
},
{
"sourceEventId": "source_guide_observation",
"relation": "supports",
"paths": [
"/details/reference/locator",
"/details/reference/sourceRevision",
"/details/reference/availability"
],
"verification": {
"method": "manual_check",
"result": "file_observed",
"artifact": "source_guide_observation",
"revision": "example_guide_r2",
"environment": "Authorized synthetic project checkout",
"coverage": "Presence and observed revision only",
"limitations": [
"No host discovery, loading, or command-execution verification"
]
}
}
],
"state": "active",
"conflictKey": "example_service:preview_example:verification_guide",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null,
"recheckWhen": [
"Guide changes",
"Project guidance identifies a different resource",
"Host or adapter changes"
]
},
"supersedes": []
},
{
"schemaVersion": 1,
"id": "memory_typed_experiment",
"revision": 1,
"kind": "working_continuity",
"facet": "verification",
"assertionType": "temporary_state",
"scope": {
"profileId": "example_measurement",
"projectId": "training_example",
"taskId": "trial_example"
},
"claim": "The baseline has a completed measurement; the candidate crashed and has no valid score.",
"rationale": "An exported numeric placeholder must not be ranked as an evaluated result.",
"futureAction": "Keep the baseline observation and the candidate failure separate; inspect the OOM before a further authorized experiment.",
"applicability": {
"taskId": "trial_example"
},
"exceptions": [],
"details": {
"experiment": {
"objective": {
"metric": "validation_loss",
"direction": "lower"
},
"evaluator": {
"revision": "eval_v1"
},
"dataset": {
"revision": "data_v1"
},
"environment": {
"hardware": "GPU_A",
"seed": 7
},
"budget": {
"value": 300,
"unit": "seconds",
"includes": "training",
"excludes": [
"startup",
"compilation"
]
},
"runs": [
{
"id": "baseline",
"codeRevision": "train_r1",
"status": "completed",
"score": 1.2,
"disposition": "pending",
"artifact": "source_typed_experiment_runs"
},
{
"id": "candidate",
"codeRevision": "train_r2",
"status": "crashed",
"score": null,
"failure": "OOM before evaluation",
"disposition": "pending",
"artifact": "source_typed_experiment_runs"
}
],
"result": "One completed baseline; no candidate comparison is possible.",
"limitations": [
"Single baseline run",
"No uncertainty estimate",
"No successful candidate evaluation",
"No adopted architecture decision"
]
}
},
"evidenceClass": "observed_verified",
"evidence": [
{
"sourceEventId": "source_experiment_objective",
"relation": "supports",
"paths": [
"/details/experiment/objective",
"/details/experiment/evaluator",
"/details/experiment/dataset",
"/details/experiment/budget",
"/details/experiment/limitations/3"
]
},
{
"sourceEventId": "source_typed_experiment_runs",
"relation": "supports",
"paths": [
"/claim",
"/details/experiment/environment",
"/details/experiment/runs",
"/details/experiment/result",
"/details/experiment/limitations/0",
"/details/experiment/limitations/1",
"/details/experiment/limitations/2"
],
"verification": {
"method": "benchmark",
"result": "baseline_completed_candidate_crashed",
"artifact": "source_typed_experiment_runs",
"revision": "train_r1 and train_r2",
"environment": "GPU_A; eval_v1; data_v1; seed 7; 300 training seconds",
"coverage": "One baseline observation and one failed run",
"limitations": [
"No valid candidate score",
"No proven comparative improvement"
]
}
}
],
"state": "active",
"conflictKey": "example_measurement:trial_example:state",
"publication": "private",
"validity": {
"validFrom": null,
"validUntil": null,
"lastVerifiedAt": null,
"recheckWhen": [
"New run available",
"Evaluator, dataset, hardware, seed, or budget changes",
"Task completes"
]
},
"supersedes": []
}
],
"pairing_examples": [
{
"id": "matched_feedback_styles",
"currentTask": "Fix the reproducible duplicate event delivery race in queue fixture Q. Preserve the ownership contract and verify the fix.",
"fixedControls": [
"Repository revision",
"Requirements",
"Engine/model",
"Inference budget",
"Tool environment"
],
"variants": [
{
"profileId": "example_a",
"memoryIds": [
"memory_style_a"
],
"expectedFirstMove": "Explain the affected state transitions and ownership invariant before constructing the reproducer."
},
{
"profileId": "example_b",
"memoryIds": [
"memory_style_b"
],
"expectedFirstMove": "Construct the minimal failing reproducer before the bounded fix and correctness explanation."
}
],
"sharedAcceptance": [
"Correct repair",
"Meaningful regression checks",
"No invented preference outside the stated conditions",
"No cross-profile memory access"
]
},
{
"id": "changed_decision_condition",
"profileId": "example_service",
"currentTask": "The preview server now has four independent workers. Ensure cached content is refreshed consistently after an edit.",
"recallFragment": {
"memoryId": "memory_cache_decision",
"revision": 1,
"disposition": "historical_decision_to_reconsider",
"selectionReason": "Same component; the current requirement matches the decision's explicit revisit condition.",
"caution": "Single-process applicability no longer holds. Neither a distributed cache nor any particular vendor is automatically required."
},
"expectedBehavior": "Inspect the current invalidation contract, compare suitable coordination options, and explain how the new worker topology changes the original tradeoff. Do not infer that the developer now prefers Redis globally."
},
{
"id": "honest_investigation_handoff",
"profileId": "example_measurement",
"currentTask": "Continue the search latency investigation in the other framework.",
"recallFragment": {
"memoryId": "memory_open_experiment",
"revision": 1,
"disposition": "investigation_lead",
"selectionReason": "Same unfinished task.",
"caution": "Active memory status records the investigation state; it does not confirm the cache hypothesis."
},
"expectedBehavior": "Preserve the observation and its confounds, check the current setup, and propose or run an authorized controlled comparison. Do not report a proven twofold cache speedup."
},
{
"id": "current_workspace_guidance",
"profileId": "example_service",
"currentTask": "Verify a preview renderer change in the other framework. The canonical guide now has revision example_guide_r3.",
"recallFragment": {
"memoryId": "memory_preview_guide",
"revision": 1,
"disposition": "resource_pointer_to_refresh",
"selectionReason": "Same project component and verification task.",
"caution": "Observed guide revision and host-delivery state may be stale."
},
"expectedBehavior": "Read current authorized project guidance, update the reference observation, and use the applicable checks. Do not claim a saved pointer proves that the other host loaded the guide."
},
{
"id": "failed_experiment_handoff",
"profileId": "example_measurement",
"currentTask": "Continue the training experiment in the other framework.",
"recallFragment": {
"memoryId": "memory_typed_experiment",
"revision": 1,
"disposition": "investigation_lead",
"selectionReason": "Same unfinished experiment.",
"caution": "The candidate has no valid performance score and there is no adopted design decision."
},
"expectedBehavior": "Preserve baseline and failure evidence, check present evaluator and resource conditions, and avoid ranking a failed run by its imported numeric placeholder."
}
],
"derived_topic_examples": [
{
"id": "topic_preview_cache",
"status": "synthetic_projection",
"profileId": "example_service",
"topic": "Why the preview server uses a local cache",
"revision": 1,
"dependencies": [
{
"memoryId": "memory_cache_decision",
"revision": 1
}
],
"statements": [
{
"text": "The single-process preview server chose an in-process cache to avoid operating another service.",
"supports": [
{
"memoryId": "memory_cache_decision",
"revision": 1,
"paths": [
"/claim",
"/rationale"
]
}
]
},
{
"text": "A sequential unchanged-input fixture invoked the renderer once for two requests; distributed invalidation remains unverified.",
"supports": [
{
"memoryId": "memory_cache_decision",
"revision": 1,
"paths": [
"/details/decision/observedOutcome",
"/details/decision/unknowns"
]
}
]
}
],
"authority": "derived_view_only",
"independentEvidence": false,
"invalidation": "A dependency correction, exclusion, supersession, or forget invalidates the page, index, and owned projections before future delivery. Regeneration rechecks revisions and access.",
"projectId": "preview_example"
}
]
}