mixdog 0.9.3 → 0.9.5

package/scripts/bench/round-r13-clientmeta.json

@@ -0,0 +1,464 @@
+{
+  "round": "13",
+  "runner": "lead",
+  "opts": {
+    "provider": "openai-oauth",
+    "model": "gpt-5.5",
+    "effort": null,
+    "fast": false,
+    "env": {
+      "MIXDOG_DATA_DIR": "C:\\Users\\tempe\\.mixdog\\bench-data"
+    }
+  },
+  "tasks": 6,
+  "completed": 6,
+  "completion_rate": 100,
+  "sessions": [
+    "sess_61152_1783057816877_1783057817120_023e2391ca2b6389040710a0f40c324d",
+    "sess_52268_1783057816891_1783057817127_5bc1c64bbe5c4a5698673bc61a98bc3b",
+    "sess_55452_1783057816916_1783057817161_5c20dccba874865f7b2d10ce85b7ef95",
+    "sess_49304_1783057907855_1783057908084_7050982a0db72b0dcbdfe2f43a649c34",
+    "sess_59336_1783057927217_1783057927463_8079787653b69fd7bc75a898880ebfca",
+    "sess_24180_1783058017094_1783058017348_05fca0c0bad5d3e560120be956dac063"
+  ],
+  "results": [
+    {
+      "id": "r4-debug-forwarding-gap#1",
+      "agent": "reviewer",
+      "ok": true,
+      "ms": 110340,
+      "sessionId": "sess_61152_1783057816877_1783057817120_023e2391ca2b6389040710a0f40c324d",
+      "output": "sessionId: sess_61152_1783057816877_1783057817120_023e2391ca2b6389040710a0f40c324d\n조사 범위를 활성화 진입점과 첫 전송 경로까지 한 번에 좁혀 확인하겠습니다.조사 완료: `bridge-state.json` 기록 뒤 첫 Discord forward까지의 경로에는 “기존 파일을 EOF로 바인딩”하는 경우가 가장 위험하고, missing transcript/watch/ownership 창은 대체로 backlog catch-up이 있습니다.\n\n**경로 요약**  \n`start()`가 `channelBridgeActive=true`와 `bridge-state.json`을 기록한 뒤 `refreshBridgeOwnership()` → `startOwnedRuntime()` → `bindPersistedTranscriptIfAny()` → `rebindTranscriptContext()` → `applyTranscriptBinding()` → `OutputForwarder.startWatch()` → `forwardNewText()` → `backend.sendMessage()`로 이어집니다. 주요 앵커는 `src/runtime/channels/index.mjs:1734`, `src/runtime/channels/index.mjs:710`, `src/runtime/channels/index.mjs:577`, `src/runtime/channels/index.mjs:394`, `src/runtime/channels/index.mjs:317`, `src/runtime/channels/lib/output-forwarder.mjs:647`, `src/runtime/channels/lib/output-forwarder.mjs:351`, `src/runtime/channels/index.mjs:284`입니다.\n\n**1. 기존 transcript를 EOF로 바인딩하면 이미 쓰인 assistant 출력은 복구되지 않습니다.**  \n`OutputForwarder.setContext()`는 `replayFromStart`도, 같은 transcript의 `catchUpFromPersisted`도 아니면 `readFileSize/lastFileSize`를 현재 파일 크기로 둡니다 (`src/runtime/channels/lib/output-forwarder.mjs:115`, `src/runtime/channels/lib/output-forwarder.mjs:117`, `src/runtime/channels/lib/output-forwarder.mjs:125`). 활성화 중 assistant 출력이 먼저 파일에 들어간 뒤 이 경로로 바인딩되면 그 bytes는 “이미 읽은 것”이 되어 backlog catch-up으로 복구되지 않습니다.\n\n**2. persisted cursor가 다른 transcript에 속하면 catch-up이 아니라 EOF skip이 됩니다.**  \n부팅/활성화 복원은 `catchUpFromPersisted: true`로 들어가지만 (`src/runtime/channels/index.mjs:625`), `sameResolvedPath()`가 아니면 `setContext()`가 현재 크기를 선택합니다 (`src/runtime/channels/lib/output-forwarder.mjs:120`, `src/runtime/channels/lib/output-forwarder.mjs:123`). 즉 새 transcript가 이미 존재하는 상태로 발견되면 activation-window 출력은 조용히 스킵되고, 이후 catch-up은 그 앞부분을 되살리지 못합니다.\n\n**3. channel binding이 없으면 자동 복원 경로가 끊기고, 나중에 바인딩돼도 기존 backlog는 보장되지 않습니다.**  \n`bindPersistedTranscriptIfAny()`는 status/main channel을 못 찾으면 그냥 반환합니다 (`src/runtime/channels/index.mjs:610`, `src/runtime/channels/index.mjs:624`), `forwardNewText()`도 `channelId` 없으면 반환합니다 (`src/runtime/channels/lib/output-forwarder.mjs:352`). 첫 inbound에서 바인딩될 수는 있지만 기존 파일 fast-path는 EOF 바인딩이어서 backlog catch-up이 아니라 skip이 될 수 있습니다 (`src/runtime/channels/index.mjs:1597`, `src/runtime/channels/index.mjs:1602`).\n\n**4. not-yet-created transcript의 `fs.watch` 실패는 보통 지연이며 60초 안에는 복구됩니다.**  \n`startWatch()`는 없는 파일에서 catch 후 watcher를 닫습니다 (`src/runtime/channels/lib/output-forwarder.mjs:656`, `src/runtime/channels/lib/output-forwarder.mjs:667`), 이를 `schedulePendingTranscriptRearm()`가 250ms polling으로 보완하고 파일 생성 시 `startWatch()`와 `forwardNewText()`를 호출합니다 (`src/runtime/channels/index.mjs:357`, `src/runtime/channels/index.mjs:376`, `src/runtime/channels/index.mjs:378`). 다만 60초 timeout 뒤 생성되면 자동 catch-up은 중단됩니다 (`src/runtime/channels/index.mjs:384`).\n\n**5. ownership/backend 연결 전 watcher flush는 drop이 아니라 지연입니다.**  \nforwarder ownership probe는 `bridgeRuntimeConnected && currentOwnerState().owned`라서 연결 전 `forwardNewText()`가 바로 false로 끝납니다 (`src/runtime/channels/index.mjs:316`, `src/runtime/channels/lib/output-forwarder.mjs:36`, `src/runtime/channels/lib/output-forwarder.mjs:351`). 이때 읽기/커밋을 하지 않으므로 bytes는 남고, 연결 후 `forwardNewText()` 또는 re-arm 재스케줄이 catch-up합니다 (`src/runtime/channels/index.mjs:801`, `src/runtime/channels/index.mjs:803`, `src/runtime/channels/index.mjs:809`).\n\n판정: activation-window의 recoverable delay는 처리되어 있지만, “이미 존재하는 transcript를 EOF로 바인딩”하는 경로는 assistant 출력이 조용히 유실될 수 있습니다.",
+      "card": {
+        "session": "sess_61152_1783057816877_1783057817120_023e2391ca2b6389040710a0f40c324d",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 99848,
+        "turns": 5,
+        "tool_calls": 6,
+        "tools_per_turn": 1.2,
+        "total_tool_ms": 562,
+        "llm_stream_ms": 98843,
+        "cache_ratio": 0.38371452563905767,
+        "cached_tokens": 42496,
+        "prompt_tokens": 110749,
+        "uncached_tokens": 68253,
+        "output_tokens": 9489,
+        "cache_weighted_input_10": 72502.6,
+        "cache_weighted_input_25": 78877,
+        "prompt_growth": 42096,
+        "antipatterns": 1,
+        "issues": 4,
+        "issue_types": [
+          "cache",
+          "broad_tool_result",
+          "service_tier",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    },
+    {
+      "id": "r4-impl-usage-summary#1",
+      "agent": "worker",
+      "ok": true,
+      "ms": 264690,
+      "sessionId": "sess_52268_1783057816891_1783057817127_5bc1c64bbe5c4a5698673bc61a98bc3b",
+      "output": "sessionId: sess_52268_1783057816891_1783057817127_5bc1c64bbe5c4a5698673bc61a98bc3b\n요청 범위가 명확해 바로 구현하고 검증까지 진행하겠습니다.`usage:today`가 이미 일부 존재해 요구사항과 맞는지 확인한 뒤 필요한 보완만 하겠습니다.필요한 구현은 `package.json`과 새 스크립트에만 한정되어 있어 바로 검증합니다.검증은 통과했고, 최종 변경 범위만 확인합니다.재영님, `usage:today` 기능 구현과 검증을 완료했습니다.\n\n- **변경 요약**: `package.json`에 `usage:today` 스크립트를 추가하고, `scripts/usage-today.mjs`를 새로 추가했습니다.\n\n- **동작**: `MIXDOG_DATA_DIR` 또는 `~/.mixdog/data/history/agent-trace.jsonl`을 읽어 오늘 자정 이후 usage를 모델별/총합으로 집계합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs`와 `npm run usage:today` 모두 성공했고, 실행 결과 `GRAND TOTAL`까지 출력됐습니다.\n\n- **참고**: 기존 작업 중인 변경으로 보이는 `src/runtime/channels/lib/*` 파일들은 건드리지 않았습니다.",
+      "card": {
+        "session": "sess_52268_1783057816891_1783057817127_5bc1c64bbe5c4a5698673bc61a98bc3b",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 254191,
+        "turns": 9,
+        "tool_calls": 10,
+        "tools_per_turn": 1.1,
+        "total_tool_ms": 2612,
+        "llm_stream_ms": 62036,
+        "cache_ratio": 0.7420684003247764,
+        "cached_tokens": 97792,
+        "prompt_tokens": 131783,
+        "uncached_tokens": 33991,
+        "output_tokens": 4789,
+        "cache_weighted_input_10": 43770.2,
+        "cache_weighted_input_25": 58439,
+        "prompt_growth": 17331,
+        "antipatterns": 1,
+        "issues": 4,
+        "issue_types": [
+          "cache",
+          "service_tier",
+          "missed_parallelism",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    },
+    {
+      "id": "r4-explain-compact-engine#1",
+      "agent": "explore",
+      "ok": true,
+      "ms": 90941,
+      "sessionId": "sess_55452_1783057816916_1783057817161_5c20dccba874865f7b2d10ce85b7ef95",
+      "output": "sessionId: sess_55452_1783057816916_1783057817161_5c20dccba874865f7b2d10ce85b7ef95\n완료했습니다, 재영님. compaction subsystem 아키텍처 맵은 아래와 같습니다.\n\n- **엔진 진입점**: `semanticCompactMessages`와 `recallFastTrackCompactMessages`가 `engine.mjs`의 공개 엔트리이며 `compact.mjs`가 재수출합니다 (`engine.mjs:278`, `engine.mjs:445`, `compact.mjs:52-55`).\n- **주 호출자**: pre-send loop는 정책 산출 뒤 recall-fasttrack 또는 semantic 엔진을 호출하고, manager/manual 경로는 `runSessionCompaction`을 호출합니다 (`loop.mjs:329-364`, `loop.mjs:414-459`, `manager.mjs:1849`, `manager.mjs:1950`).\n- **manual/auto-clear runner**: `runSessionCompaction`은 같은 엔진을 import해 type별로 recall-fasttrack, semantic, prune fallback을 수행합니다 (`compaction-runner.mjs:8-16`, `compaction-runner.mjs:143-147`, `compaction-runner.mjs:207-306`).\n- **recall pipeline 래퍼**: loop의 `runRecallFastTrackCompact`가 memory ingest/dump/cycle1 후 `recallFastTrackCompactMessages`에 recall text와 tail 정책을 넘깁니다 (`recall-fasttrack.mjs:61-84`, `recall-fasttrack.mjs:103-148`, `recall-fasttrack.mjs:161-172`).\n\n- **자동 발화 조건**: loop는 queued steering을 먼저 반영한 뒤 정책을 만들고 현재 messages token estimate로 compaction 여부를 판단합니다 (`loop.mjs:324-353`).\n- **정책 산출**: auto가 꺼지면 중단하고, boundary는 context/boundary/autoLimit에서 산출하며, explicit autoLimit은 boundary보다 작을 때만 trigger가 됩니다 (`compact-policy.mjs:90-117`).\n- **압력 계산**: compaction은 reactive overflow면 무조건, token estimate가 없으면 보수적으로, 아니면 `messages + reserve >= trigger`일 때 발화합니다 (`compact-policy.mjs:151-174`).\n- **대상 예산**: compact target은 boundary 안에서 targetEffective + request/configured reserve로 잡습니다 (`compact-policy.mjs:118-148`, `compact-policy.mjs:163-168`, `loop.mjs:362-364`).\n- **manual runner 조건**: runner는 auto disabled, 짧은 history, context window 없음, below-threshold를 조기 반환하고 manual/force는 threshold를 우회합니다 (`compaction-runner.mjs:143-199`).\n\n- **공통 분리 단계**: 엔진은 tool pair sanitize/dedup/reconcile 후 protected prefix, previous summary, live(non-summary)를 분리합니다 (`engine.mjs:100-111`, `messages.mjs:56-81`).\n- **semantic tail 선택**: live에 user turn이 있으면 기본 2개 최신 turn을 budget 안에서 보존하고, 최신 turn이 너무 크면 suffix를 잘라 보존합니다 (`engine.mjs:46`, `engine.mjs:134-185`, `engine.mjs:235-245`).\n- **no-user semantic tail**: user turn이 없으면 tool result로 시작하지 않는 cut point만 허용해 assistant/tool-only transcript도 provider-valid tail로 유지합니다 (`engine.mjs:121-127`, `engine.mjs:193-232`).\n- **tail 후처리**: 선택된 tail은 다시 sanitize/dedup/reconcile되며, empty live만 unrecoverable입니다 (`engine.mjs:244-255`).\n- **recall tail 정책**: recall-fasttrack은 최신 최대 2 user-anchored turns, 기본 8k cap을 보존하고 초과 시 오래된 turn drop 후 truncation합니다 (`engine.mjs:449-468`, `engine.mjs:593-611`).\n- **recall no-user tail**: recall tail도 no-user 구조를 허용하고, tool orphan 방지를 위해 backward 확장 및 sanitize guard를 둡니다 (`engine.mjs:525-575`, `engine.mjs:613-619`).\n\n- **redaction 시점**: fits-budget no-op은 기존 semantics 유지를 위해 redaction 없이 sanitized transcript를 반환합니다 (`engine.mjs:284-304`, `engine.mjs:625-643`).\n- **verbatim tail redaction**: compaction이 실제 진행될 때만 window selection 전에 tool-call secret argument values를 redact해 system/tail이 redacted 상태로 측정·출력됩니다 (`engine.mjs:306-314`, `engine.mjs:389-392`, `engine.mjs:644-646`).\n- **redaction 구현**: redactor는 assistant toolCalls argument values만 shape/order/pairing 보존 방식으로 바꾸며 raw key/value secrets도 제거합니다 (`text-utils.mjs:217-326`).\n- **prompt-side redaction**: summarized head는 prompt transcript 생성 시 tool-call summary와 argument normalization 경로에서 별도로 truncation/redaction됩니다 (`summary.mjs:85-115`, `text-utils.mjs:121-160`).\n\n- **semantic 실패 지점**: provider 없음, compactable head 없음, prompt fit 실패, empty summary, summary fit 실패, final budget 초과가 throw됩니다 (`engine.mjs:279-281`, `engine.mjs:315-337`, `engine.mjs:371-386`, `engine.mjs:395-397`).\n- **recall 실패 지점**: compactable prior/recall text 없음, recall summary fit 실패, final budget 초과가 throw됩니다 (`engine.mjs:653-668`, `engine.mjs:691-700`).\n- **budget fallback**: mandatory system+tail이 target보다 크면 거절하지 않고 summary room만큼 budget을 raise합니다 (`engine.mjs:319-330`, `engine.mjs:658-663`).\n- **schema fallback**: semantic provider response가 malformed여도 summary schema를 deterministic repair 후 사용합니다 (`engine.mjs:373-377`).\n- **loop fallback**: semantic 실패는 non-LLM `pruneToolOutputsUnanchored`로 한 번 salvage하고, recall-fasttrack 실패는 prune으로 숨기지 않고 overflow failure로 escalate합니다 (`loop.mjs:486-536`, `loop.mjs:546-626`).\n- **runner fallback**: manager/manual 경로는 recall-fasttrack 실패 시 semantic fallback을 한 번 시도하고, non-recall 실패에는 prune fallback을 시도한 뒤 실패 metadata를 기록합니다 (`compaction-runner.mjs:231-278`, `compaction-runner.mjs:322-397`).\n- **성공 적용**: 성공 시 transcript를 compacted messages로 교체하고 providerState를 무효화하며 compaction telemetry를 갱신합니다 (`loop.mjs:629-664`, `compaction-runner.mjs:399-470`).",
+      "card": {
+        "session": "sess_55452_1783057816916_1783057817161_5c20dccba874865f7b2d10ce85b7ef95",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 80435,
+        "turns": 11,
+        "tool_calls": 21,
+        "tools_per_turn": 1.9,
+        "total_tool_ms": 849,
+        "llm_stream_ms": 79479,
+        "cache_ratio": 0.7835232577551613,
+        "cached_tokens": 253440,
+        "prompt_tokens": 323462,
+        "uncached_tokens": 70022,
+        "output_tokens": 5788,
+        "cache_weighted_input_10": 95366,
+        "cache_weighted_input_25": 133382,
+        "prompt_growth": 51254,
+        "antipatterns": 1,
+        "issues": 6,
+        "issue_types": [
+          "cache_miss",
+          "readonly_stall",
+          "cache",
+          "service_tier",
+          "missed_parallelism",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    },
+    {
+      "id": "r4-debug-forwarding-gap#2",
+      "agent": "reviewer",
+      "ok": true,
+      "ms": 123581,
+      "sessionId": "sess_49304_1783057907855_1783057908084_7050982a0db72b0dcbdfe2f43a649c34",
+      "output": "sessionId: sess_49304_1783057907855_1783057908084_7050982a0db72b0dcbdfe2f43a649c34\n완료: `bridge-state.json` 기록부터 첫 전송까지 추적했고, 핵심 리스크는 “전송 준비 전 기존 파일 크기를 EOF로 잡는 바인딩”입니다.\n\n**경로 요약**: activation은 `src/runtime/channels/lib/tool-dispatch.mjs:45-64` 또는 `src/runtime/channels/index.mjs:1734-1747`에서 state 기록 → ownership/start runtime `src/runtime/channels/index.mjs:881-927`, `710-813` → transcript bind `577-629`, `394-516`, `317-337` → watch/flush/send `src/runtime/channels/lib/output-forwarder.mjs:647-731`, `351-410`, `283-340`.\n\n**1. 이미 쓰인 transcript를 첫 바인딩할 때 EOF로 커서가 잡힘**: `setContext()`는 `replayFromStart`도 아니고 동일 persisted cursor도 아니면 `currentSize`를 시작점으로 삼아, bridge 활성화 후 바인딩 전 생성된 assistant bytes를 “과거분”으로 보고 건너뜁니다. 복구: 기본 backlog catch-up으로는 복구 안 됨; 같은 transcript의 `lastFileSize`가 persisted되어 있거나 pending 경로가 `replayFromStart`로 잡힌 경우만 복구됩니다. `src/runtime/channels/lib/output-forwarder.mjs:111-127`, `src/runtime/channels/index.mjs:625-629`, `433-449`.\n\n**2. transcript가 아직 없어서 `fs.watch` 설치가 실패하는 구간**: pending path에 bind하면 `startWatch()`가 `ENOENT`류로 닫히고, `schedulePendingTranscriptRearm()`가 250ms polling/60s timeout으로 생성을 기다린 뒤 `startWatch()`와 `forwardNewText()`를 호출합니다. 복구: 60초 내 생성되면 backlog 복구, timeout 이후에는 다음 rebind/다른 flush 계기가 없으면 지연 상태로 남습니다. `src/runtime/channels/index.mjs:339-346`, `357-392`, `493-510`; `src/runtime/channels/lib/output-forwarder.mjs:647-670`.\n\n**3. ownership probe가 connect 전에는 fail-closed라 flush가 무시됨**: ownerGetter는 `bridgeRuntimeConnected && currentOwnerState().owned`라서 backend connect 전 `forwardNewText()`/`forwardFinalText()`는 false/return으로 끝나지만 커서는 전진하지 않습니다. 복구: post-connect에서 pending rearm 재설치 또는 `forwardNewText()`를 호출하므로 정상적으로는 지연만 발생합니다. `src/runtime/channels/index.mjs:316`, `758-763`, `776`, `801-812`; `src/runtime/channels/lib/output-forwarder.mjs:351-353`, `490-497`.\n\n**4. watch 설치 직전/이벤트 누락 구간**: `startWatch()`는 설치 직후 `scheduleWatchFlush()`를 강제 호출하므로 stat 이후 watch 설치 전 쓰인 bytes나 누락된 첫 이벤트는 100ms debounce 뒤 읽힙니다. 복구: backlog catch-up 됨; 최악은 첫 전송 지연입니다. `src/runtime/channels/lib/output-forwarder.mjs:665-666`, `703-731`.\n\n**5. target channel이 없으면 bind 자체가 중단됨**: activation restore는 status에서 channel을 찾고 없으면 main channel로 자동 bind하지만, 둘 다 없으면 반환되어 output forwarding 경로가 없습니다. 복구: 이후 inbound로 channel이 생겨 rebind되면 가능하지만, 그때 EOF 바인딩이면 기존 bytes는 다시 1번 리스크를 탑니다. `src/runtime/channels/index.mjs:577-624`.\n\n판정: bridge activation은 대부분 지연 후 catch-up되지만, “바인딩 전 이미 존재한 bytes를 EOF 기준으로 시작하는 경우”는 실제 silent drop 가능성이 남아 있습니다.",
+      "card": {
+        "session": "sess_49304_1783057907855_1783057908084_7050982a0db72b0dcbdfe2f43a649c34",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 113090,
+        "turns": 9,
+        "tool_calls": 13,
+        "tools_per_turn": 1.4,
+        "total_tool_ms": 932,
+        "llm_stream_ms": 112197,
+        "cache_ratio": 0.6753989684387698,
+        "cached_tokens": 261632,
+        "prompt_tokens": 387374,
+        "uncached_tokens": 125742,
+        "output_tokens": 9822,
+        "cache_weighted_input_10": 151905.2,
+        "cache_weighted_input_25": 191150,
+        "prompt_growth": 56276,
+        "antipatterns": 1,
+        "issues": 8,
+        "issue_types": [
+          "cache_miss",
+          "readonly_stall",
+          "cache",
+          "broad_tool_result",
+          "service_tier",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    },
+    {
+      "id": "r4-impl-usage-summary#2",
+      "agent": "worker",
+      "ok": true,
+      "ms": 89816,
+      "sessionId": "sess_59336_1783057927217_1783057927463_8079787653b69fd7bc75a898880ebfca",
+      "output": "sessionId: sess_59336_1783057927217_1783057927463_8079787653b69fd7bc75a898880ebfca\n요청 범위에 맞춰 기존 스크립트 스타일과 package.json만 확인한 뒤 바로 구현하겠습니다.현재 브랜치에는 `scripts/usage-today.mjs`와 `usage:today`가 이미 있어, 요구사항 충족 여부를 맞춰 확인하고 필요한 보정만 하겠습니다.요구사항 대부분이 이미 반영된 상태라, 실제 변경 여부를 `git` 기준으로 확인하고 부족한 부분만 패치하겠습니다.기능 코드는 요구사항과 맞지만, trace 포맷 호환성만 한 번 더 맞춘 뒤 검증하겠습니다.필요한 항목이 이미 target diff에 잡혀 있어, 문법 검사와 실제 실행으로 확인하겠습니다.검증은 통과했고, 최종 보고용으로 target diff만 확인하겠습니다.재영님, `usage:today` 기능 구현 및 검증 완료했습니다.\n\n- **Diff 요약**: `package.json`에 `usage:today` npm script 추가, `scripts/usage-today.mjs` 신규 추가.\n\n- **동작**: `MIXDOG_DATA_DIR` 또는 `~/.mixdog/data/history/agent-trace.jsonl`을 읽고, 로컬 자정 이후 usage row를 모델별로 집계해 `calls/input/cached/output/costUsd`와 `GRAND TOTAL`을 출력합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 성공.\n\n- **참고**: 현재 워킹트리에 별도 변경 `src/runtime/channels/lib/event-queue.mjs`, `src/runtime/channels/lib/output-forwarder.mjs`가 남아 있습니다.",
+      "card": {
+        "session": "sess_59336_1783057927217_1783057927463_8079787653b69fd7bc75a898880ebfca",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 79319,
+        "turns": 10,
+        "tool_calls": 13,
+        "tools_per_turn": 1.3,
+        "total_tool_ms": 2980,
+        "llm_stream_ms": 76320,
+        "cache_ratio": 0.5524303168138784,
+        "cached_tokens": 104448,
+        "prompt_tokens": 189070,
+        "uncached_tokens": 84622,
+        "output_tokens": 4364,
+        "cache_weighted_input_10": 95066.8,
+        "cache_weighted_input_25": 110734,
+        "prompt_growth": 24627,
+        "antipatterns": 1,
+        "issues": 6,
+        "issue_types": [
+          "cache_miss",
+          "cache",
+          "service_tier",
+          "missed_parallelism",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    },
+    {
+      "id": "r4-explain-compact-engine#2",
+      "agent": "explore",
+      "ok": true,
+      "ms": 85751,
+      "sessionId": "sess_24180_1783058017094_1783058017348_05fca0c0bad5d3e560120be956dac063",
+      "output": "sessionId: sess_24180_1783058017094_1783058017348_05fca0c0bad5d3e560120be956dac063\n- **Entry points/callers:** `engine.mjs` exposes `semanticCompactMessages` and `recallFastTrackCompactMessages`; `compact.mjs` re-exports them for existing callers (`loop.mjs`, `manager.mjs`, session-builder, memory, scripts). `src/runtime/agent/orchestrator/session/compact/engine.mjs:278`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:445`, `src/runtime/agent/orchestrator/session/compact.mjs:1`, `src/runtime/agent/orchestrator/session/compact.mjs:52`\n\n- **Loop trigger:** Before each provider send, the loop drains queued steering, resolves policy, estimates tokens, and compacts when `forceReactive` is set, token estimate is unknown, or transcript pressure reaches `triggerTokens`. `src/runtime/agent/orchestrator/session/loop.mjs:324`, `src/runtime/agent/orchestrator/session/loop.mjs:329`, `src/runtime/agent/orchestrator/session/loop.mjs:351`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`\n\n- **Policy math:** Auto compact is enabled unless disabled, boundary comes from context/window metadata, trigger is explicit sub-boundary limit or `boundary - buffer`, reserve includes tool/schema reserve, and type resolves to semantic or recall-fasttrack. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:101`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:107`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:117`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:121`\n\n- **Manual/session runner:** `runSessionCompaction` handles manual/auto session compaction; auto exits when disabled, too small, missing boundary, or below threshold, while manual forces compaction. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:143`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:145`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:147`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:149`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:176`\n\n- **Semantic path:** Semantic compaction sanitizes first, no-ops if already within budget and not forced, then redacts tool-call secrets before selecting the window and building the summary prompt. `src/runtime/agent/orchestrator/session/compact/engine.mjs:283`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:289`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:306`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:312`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:314`\n\n- **Preserved tail:** Windowing splits protected/system context, previous summary, and live non-summary messages; user-turn sessions keep up to `tailTurns` recent turns within `preserveRecentBudget`, while no-user sessions choose the newest valid non-tool cut that fits. `src/runtime/agent/orchestrator/session/compact/engine.mjs:100`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:110`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:235`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:240`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`\n\n- **Tool-pair safety:** Tail boundaries never start on a `tool` message, and the chosen tail is re-sanitized/deduped/reconciled so assistant tool calls and tool results remain provider-valid. `src/runtime/agent/orchestrator/session/compact/engine.mjs:121`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:124`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:246`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:250`\n\n- **Redaction hook:** Secret keys/raw `key:value` strings are normalized in `text-utils`, and semantic/recall paths both call `redactToolCallSecretsInMessages` after base sanitization so emitted preserved system/tail content is redacted. `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:12`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:46`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:312`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:644`\n\n- **Recall-fasttrack path:** Recall pipeline ingests session history, dumps memory roots, drains raw cycle1 rows when needed, caps recall injection to 5% of context with a floor, then calls `recallFastTrackCompactMessages`. `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:61`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:84`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:109`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:151`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:161`\n\n- **Recall tail:** Recall keeps the newest user-anchored turns, drops oldest turns if over cap, truncates the oldest surviving content if needed, and allows no-user assistant/tool tails. `src/runtime/agent/orchestrator/session/compact/engine.mjs:593`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:601`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:602`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:610`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:613`\n\n- **Budget/fallback:** If mandatory system+tail plus summary room exceeds target, semantic/recall raise the working budget instead of refusing; if semantic fails, loop/runner try unanchored tool-output prune, but recall-fasttrack failures are not silently pruned. `src/runtime/agent/orchestrator/session/compact/engine.mjs:319`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:328`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:658`, `src/runtime/agent/orchestrator/session/loop.mjs:501`, `src/runtime/agent/orchestrator/session/loop.mjs:511`\n\n- **Failure paths:** Recall-fasttrack runner can fall back once to semantic in `runSessionCompaction`; loop failures record `overflow_failed`, emit failure telemetry, and throw `AgentContextOverflowError`; provider send overflow gets one reactive compact retry. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:237`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:245`, `src/runtime/agent/orchestrator/session/loop.mjs:552`, `src/runtime/agent/orchestrator/session/loop.mjs:618`, `src/runtime/agent/orchestrator/session/loop.mjs:982`",
+      "card": {
+        "session": "sess_24180_1783058017094_1783058017348_05fca0c0bad5d3e560120be956dac063",
+        "provider": "openai-oauth",
+        "model": "gpt-5.5",
+        "wall_ms": 75181,
+        "turns": 9,
+        "tool_calls": 8,
+        "tools_per_turn": 0.9,
+        "total_tool_ms": 1343,
+        "llm_stream_ms": 66031,
+        "cache_ratio": 0.5456843130985551,
+        "cached_tokens": 107520,
+        "prompt_tokens": 197037,
+        "uncached_tokens": 89517,
+        "output_tokens": 4375,
+        "cache_weighted_input_10": 100269,
+        "cache_weighted_input_25": 116397,
+        "prompt_growth": 47535,
+        "antipatterns": 1,
+        "issues": 8,
+        "issue_types": [
+          "cache_miss",
+          "readonly_stall",
+          "cache",
+          "broad_tool_result",
+          "service_tier",
+          "missed_parallelism",
+          "tool_churn_cluster"
+        ],
+        "read_fragmentation_paths": [],
+        "grep_sweep_count": 0
+      }
+    }
+  ],
+  "task_complete": true,
+  "task_errors": [],
+  "score_complete": true,
+  "score_errors": [],
+  "group": {
+    "n": 6,
+    "wall_ms": 117010.7,
+    "turns": 8.8,
+    "tool_calls": 11.8,
+    "tools_per_turn": 1.3,
+    "total_tool_ms": 1546.3,
+    "llm_stream_ms": 82484.3,
+    "cache_ratio": 0.6,
+    "cached_tokens": 144554.7,
+    "prompt_tokens": 223245.8,
+    "uncached_tokens": 78691.2,
+    "output_tokens": 6437.8,
+    "cache_weighted_input_10": 93146.6,
+    "cache_weighted_input_25": 114829.8,
+    "prompt_growth": 39853.2,
+    "antipatterns": 1,
+    "issues": 6,
+    "grep_sweep_count": 0
+  },
+  "cards": [
+    {
+      "session": "sess_61152_1783057816877_1783057817120_023e2391ca2b6389040710a0f40c324d",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 99848,
+      "turns": 5,
+      "tool_calls": 6,
+      "tools_per_turn": 1.2,
+      "total_tool_ms": 562,
+      "llm_stream_ms": 98843,
+      "cache_ratio": 0.38371452563905767,
+      "cached_tokens": 42496,
+      "prompt_tokens": 110749,
+      "uncached_tokens": 68253,
+      "output_tokens": 9489,
+      "cache_weighted_input_10": 72502.6,
+      "cache_weighted_input_25": 78877,
+      "prompt_growth": 42096,
+      "antipatterns": 1,
+      "issues": 4,
+      "issue_types": [
+        "cache",
+        "broad_tool_result",
+        "service_tier",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    },
+    {
+      "session": "sess_52268_1783057816891_1783057817127_5bc1c64bbe5c4a5698673bc61a98bc3b",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 254191,
+      "turns": 9,
+      "tool_calls": 10,
+      "tools_per_turn": 1.1,
+      "total_tool_ms": 2612,
+      "llm_stream_ms": 62036,
+      "cache_ratio": 0.7420684003247764,
+      "cached_tokens": 97792,
+      "prompt_tokens": 131783,
+      "uncached_tokens": 33991,
+      "output_tokens": 4789,
+      "cache_weighted_input_10": 43770.2,
+      "cache_weighted_input_25": 58439,
+      "prompt_growth": 17331,
+      "antipatterns": 1,
+      "issues": 4,
+      "issue_types": [
+        "cache",
+        "service_tier",
+        "missed_parallelism",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    },
+    {
+      "session": "sess_55452_1783057816916_1783057817161_5c20dccba874865f7b2d10ce85b7ef95",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 80435,
+      "turns": 11,
+      "tool_calls": 21,
+      "tools_per_turn": 1.9,
+      "total_tool_ms": 849,
+      "llm_stream_ms": 79479,
+      "cache_ratio": 0.7835232577551613,
+      "cached_tokens": 253440,
+      "prompt_tokens": 323462,
+      "uncached_tokens": 70022,
+      "output_tokens": 5788,
+      "cache_weighted_input_10": 95366,
+      "cache_weighted_input_25": 133382,
+      "prompt_growth": 51254,
+      "antipatterns": 1,
+      "issues": 6,
+      "issue_types": [
+        "cache_miss",
+        "readonly_stall",
+        "cache",
+        "service_tier",
+        "missed_parallelism",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    },
+    {
+      "session": "sess_49304_1783057907855_1783057908084_7050982a0db72b0dcbdfe2f43a649c34",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 113090,
+      "turns": 9,
+      "tool_calls": 13,
+      "tools_per_turn": 1.4,
+      "total_tool_ms": 932,
+      "llm_stream_ms": 112197,
+      "cache_ratio": 0.6753989684387698,
+      "cached_tokens": 261632,
+      "prompt_tokens": 387374,
+      "uncached_tokens": 125742,
+      "output_tokens": 9822,
+      "cache_weighted_input_10": 151905.2,
+      "cache_weighted_input_25": 191150,
+      "prompt_growth": 56276,
+      "antipatterns": 1,
+      "issues": 8,
+      "issue_types": [
+        "cache_miss",
+        "readonly_stall",
+        "cache",
+        "broad_tool_result",
+        "service_tier",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    },
+    {
+      "session": "sess_59336_1783057927217_1783057927463_8079787653b69fd7bc75a898880ebfca",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 79319,
+      "turns": 10,
+      "tool_calls": 13,
+      "tools_per_turn": 1.3,
+      "total_tool_ms": 2980,
+      "llm_stream_ms": 76320,
+      "cache_ratio": 0.5524303168138784,
+      "cached_tokens": 104448,
+      "prompt_tokens": 189070,
+      "uncached_tokens": 84622,
+      "output_tokens": 4364,
+      "cache_weighted_input_10": 95066.8,
+      "cache_weighted_input_25": 110734,
+      "prompt_growth": 24627,
+      "antipatterns": 1,
+      "issues": 6,
+      "issue_types": [
+        "cache_miss",
+        "cache",
+        "service_tier",
+        "missed_parallelism",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    },
+    {
+      "session": "sess_24180_1783058017094_1783058017348_05fca0c0bad5d3e560120be956dac063",
+      "provider": "openai-oauth",
+      "model": "gpt-5.5",
+      "wall_ms": 75181,
+      "turns": 9,
+      "tool_calls": 8,
+      "tools_per_turn": 0.9,
+      "total_tool_ms": 1343,
+      "llm_stream_ms": 66031,
+      "cache_ratio": 0.5456843130985551,
+      "cached_tokens": 107520,
+      "prompt_tokens": 197037,
+      "uncached_tokens": 89517,
+      "output_tokens": 4375,
+      "cache_weighted_input_10": 100269,
+      "cache_weighted_input_25": 116397,
+      "prompt_growth": 47535,
+      "antipatterns": 1,
+      "issues": 8,
+      "issue_types": [
+        "cache_miss",
+        "readonly_stall",
+        "cache",
+        "broad_tool_result",
+        "service_tier",
+        "missed_parallelism",
+        "tool_churn_cluster"
+      ],
+      "read_fragmentation_paths": [],
+      "grep_sweep_count": 0
+    }
+  ]
+}