mixdog 0.9.17 → 0.9.19

package/scripts/bench/round-r16-sessionid.json

@@ -1,466 +0,0 @@
-{
-  "round": "16",
-  "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_58408_1783058979193_1783058979442_d47c240b8d093c927461b55a0dc2516c",
-    "sess_18552_1783058979205_1783058979453_87070f4cdd6037cd39aa2a84fc8356ac",
-    "sess_33120_1783058979218_1783058979467_17893d4613374d6fecfb2f6704eff5b2",
-    "sess_47252_1783059063096_1783059063358_bd9cf12a670c04ca4c9e3d75b7938009",
-    "sess_51160_1783059076138_1783059076364_ce7ebef63b9773c3457e0a8dbeed9dd7",
-    "sess_58804_1783059088033_1783059088258_fecef461f40816f3c4db38876a81fd3b"
-  ],
-  "results": [
-    {
-      "id": "r4-debug-forwarding-gap#1",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 83904,
-      "sessionId": "sess_58408_1783058979193_1783058979442_d47c240b8d093c927461b55a0dc2516c",
-      "output": "sessionId: sess_58408_1783058979193_1783058979442_d47c240b8d093c927461b55a0dc2516c\n조사 완료: `bridge-state.json`은 실제 전송 준비보다 먼저 쓰이며, 이 구간 출력은 대부분 “지연 후 catch-up”되지만 일부 조건에서는 조용히 누락될 수 있습니다.\n\n- **가장 큰 누락 위험 — 채널 바인딩 없음:** 활성화 시 저장된 `channelId`도 `mainChannel`도 없으면 바인딩이 중단되고, 이후 첫 inbound가 기존 transcript를 `catchUp` 없이 EOF로 바인딩해 이전 출력이 스킵될 수 있습니다. `src/runtime/channels/index.mjs:610`, `src/runtime/channels/index.mjs:624`, `src/runtime/channels/index.mjs:1597`, `src/runtime/channels/lib/output-forwarder.mjs:125` — backlog catch-up **복구 안 됨**.\n\n- **not-yet-created transcript:** `fs.watch`는 없는 파일에 실패하지만 pending path를 저장하고 re-arm poll이 파일 생성 시 watch 설치 후 `forwardNewText()`를 호출합니다. `src/runtime/channels/index.mjs:469`, `src/runtime/channels/index.mjs:486`, `src/runtime/channels/index.mjs:497`, `src/runtime/channels/index.mjs:376`, `src/runtime/channels/index.mjs:379` — 60초 안 생성되면 **복구됨**, timeout 이후는 자동 복구가 보장되지 않습니다.\n\n- **bridge-state 선행 노출:** `start()`가 `channelBridgeActive=true`와 `bridge-state.json`을 먼저 쓰고, 그 뒤 ownership/connect/binding을 기다립니다. `src/runtime/channels/index.mjs:1734`, `src/runtime/channels/index.mjs:1736`, `src/runtime/channels/index.mjs:1744`, `src/runtime/channels/index.mjs:769`, `src/runtime/channels/index.mjs:776` — transcript cursor가 전진하지 않아 post-connect `forwardNewText()`로 **지연 복구됨**.\n\n- **ownership probe 대기:** `_isOwner()`는 `bridgeRuntimeConnected && currentOwnerState().owned`가 될 때까지 false라 watch flush가 와도 읽지 않습니다. `src/runtime/channels/index.mjs:316`, `src/runtime/channels/lib/output-forwarder.mjs:351`, `src/runtime/channels/lib/output-forwarder.mjs:354` — 읽기 전 return이므로 backlog는 이후 flush/re-arm/post-connect forward로 **복구됨**.\n\n- **pre-connect re-arm 조기 종료:** pre-connect에서 pending re-arm이 먼저 돌면 `!bridgeRuntimeConnected`로 종료됩니다. 이후 connect 완료 시 missing transcript를 다시 re-arm합니다. `src/runtime/channels/index.mjs:372`, `src/runtime/channels/index.mjs:803`, `src/runtime/channels/index.mjs:807` — **지연 복구됨**.\n\n- **정상 첫 전송 경로:** 바인딩 후 watch가 설치되면 생성/변경 이벤트가 `scheduleWatchFlush()` → `forwardNewText()` → queue → `backend.sendMessage()`로 흐릅니다. `src/runtime/channels/lib/output-forwarder.mjs:647`, `src/runtime/channels/lib/output-forwarder.mjs:657`, `src/runtime/channels/lib/output-forwarder.mjs:724`, `src/runtime/channels/lib/output-forwarder.mjs:402`, `src/runtime/channels/lib/output-forwarder.mjs:446`, `src/runtime/channels/lib/output-forwarder.mjs:316`.\n\nVerdict: 현재 설계는 일반 활성화 레이스는 catch-up으로 회복하지만, “채널 미바인딩 상태에서 기존 transcript를 첫 inbound가 EOF 바인딩”하는 경우와 pending transcript 60초 timeout 이후는 조용한 누락 가능성이 남아 있습니다.",
-      "card": {
-        "session": "sess_58408_1783058979193_1783058979442_d47c240b8d093c927461b55a0dc2516c",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 73382,
-        "turns": 5,
-        "tool_calls": 5,
-        "tools_per_turn": 1,
-        "total_tool_ms": 408,
-        "llm_stream_ms": 72483,
-        "cache_ratio": 0.7143255762040293,
-        "cached_tokens": 89600,
-        "prompt_tokens": 125433,
-        "uncached_tokens": 35833,
-        "output_tokens": 6270,
-        "cache_weighted_input_10": 44793,
-        "cache_weighted_input_25": 58233,
-        "prompt_growth": 31998,
-        "antipatterns": 1,
-        "issues": 9,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "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": 108833,
-      "sessionId": "sess_18552_1783058979205_1783058979453_87070f4cdd6037cd39aa2a84fc8356ac",
-      "output": "sessionId: sess_18552_1783058979205_1783058979453_87070f4cdd6037cd39aa2a84fc8356ac\n요청 범위가 명확해서 바로 구현하고 검증까지 진행하겠습니다.기존 미완성 파일이 있어 누락된 데이터 경로 해석만 보완하고 검증하겠습니다.검증은 통과했고, 이제 변경 범위만 확인하겠습니다.완료했습니다, 재영님. `usage:today` 스크립트와 `scripts/usage-today.mjs` 집계 기능이 준비됐습니다.\n\n- **Diff 요약**: `package.json`에 `usage:today` 추가, `scripts/usage-today.mjs` 신규로 오늘 자정 이후 usage를 모델별/총합 집계합니다.\n\n- **출력 항목**: 모델별 `calls`, `input`, `cached`, `output`, `costUsd`와 `GRAND TOTAL`을 출력하며 `--trace`, `--json`도 지원합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 완료. 현재 실행 결과 총합은 `1413 calls`, `52.44M input`, `50.77M cached`, `1.65M output`, `$91.2403`입니다.\n\n참고로 작업 전부터 있던 `src/runtime/channels/lib/event-queue.mjs`, `src/runtime/channels/lib/output-forwarder.mjs` 변경은 건드리지 않았습니다.",
-      "card": {
-        "session": "sess_18552_1783058979205_1783058979453_87070f4cdd6037cd39aa2a84fc8356ac",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 98329,
-        "turns": 18,
-        "tool_calls": 24,
-        "tools_per_turn": 1.3,
-        "total_tool_ms": 3531,
-        "llm_stream_ms": 94684,
-        "cache_ratio": 0.8973703724131169,
-        "cached_tokens": 292864,
-        "prompt_tokens": 326358,
-        "uncached_tokens": 33494,
-        "output_tokens": 6700,
-        "cache_weighted_input_10": 62780.4,
-        "cache_weighted_input_25": 106710,
-        "prompt_growth": 23545,
-        "antipatterns": 3,
-        "issues": 16,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "tool_error",
-          "service_tier",
-          "duplicate_tool",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-explain-compact-engine#1",
-      "agent": "explore",
-      "ok": true,
-      "ms": 96923,
-      "sessionId": "sess_33120_1783058979218_1783058979467_17893d4613374d6fecfb2f6704eff5b2",
-      "output": "ap, anchored to `file:line`.\n\n- **Exports:** `engine.mjs` exposes `semanticCompactMessages()` and `recallFastTrackCompactMessages()` as the two compaction engines. `src/runtime/agent/orchestrator/session/compact/engine.mjs:278`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:445`\n\n- **Runtime callers:** auto pre-send compaction enters from `agentLoop()`, choosing recall-fasttrack first or semantic second. `src/runtime/agent/orchestrator/session/loop.mjs:414`, `src/runtime/agent/orchestrator/session/loop.mjs:439`\n\n- **Manual/clear callers:** manual `compact()` and pre-clear compaction route through `runSessionCompaction()`. `src/runtime/agent/orchestrator/session/manager.mjs:1849`, `src/runtime/agent/orchestrator/session/manager.mjs:1950`, `src/mixdog-session-runtime.mjs:2443`, `src/mixdog-session-runtime.mjs:2462`\n\n- **Policy gate:** auto compaction only exists when `resolveWorkerCompactPolicy()` returns `auto:true` with boundary, trigger, type, tail, keep, and reserve settings. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:124`\n\n- **Fire decision:** pre-send drains pending steering first, estimates message tokens, then fires if reactive overflow is pending or `messageTokens + reserveTokens >= triggerTokens`. `src/runtime/agent/orchestrator/session/loop.mjs:324`, `src/runtime/agent/orchestrator/session/loop.mjs:351`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`\n\n- **Budget:** compaction targets a reduced budget derived from boundary, target ratio/min/max, and reserve; loop uses that only when `shouldCompact` is true. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:66`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:163`, `src/runtime/agent/orchestrator/session/loop.mjs:362`\n\n- **Manual threshold:** manual mode is forced, while non-forced session compaction returns `below threshold` when pressure is under trigger. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:143`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:176`\n\n- **Semantic no-op:** if sanitized messages already fit and `force !== true`, semantic returns unchanged and explicitly skips preserved-tail redaction. `src/runtime/agent/orchestrator/session/compact/engine.mjs:284`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:286`\n\n- **Semantic split:** semantic separates protected system prefix, live non-summary history, and prior summary before selecting a compacted head plus preserved tail. `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`\n\n- **Semantic tail:** with user turns, it keeps up to `tailTurns` recent turns within recent budget, splitting newest suffix if needed. `src/runtime/agent/orchestrator/session/compact/engine.mjs:46`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:134`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:141`\n\n- **No-user tail:** without user anchors, it cuts only at non-`tool` messages so tool results never lead an orphaned tail. `src/runtime/agent/orchestrator/session/compact/engine.mjs:114`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:121`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`\n\n- **Tail repair:** after semantic cut, the preserved tail is repaired with `sanitizeToolPairs`, `dedupToolResultBodies`, and `reconcileDedupStubs`. `src/runtime/agent/orchestrator/session/compact/engine.mjs:246`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:250`\n\n- **Recall tail:** recall-fasttrack preserves newest structured turns verbatim, defaulting to two user-anchored turns under an 8k cap. `src/runtime/agent/orchestrator/session/compact/engine.mjs:449`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:467`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:593`\n\n- **Recall truncation:** if recall tail exceeds cap, it drops oldest kept turns, then truncates the last surviving turn/message while preserving at least newest content. `src/runtime/agent/orchestrator/session/compact/engine.mjs:601`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:609`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:525`\n\n- **Redaction hook:** once compaction proceeds, both semantic and recall redact tool-call secret arguments before window/tail selection, so emitted preserved context is already redacted. `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:644`\n\n- **Redaction behavior:** redaction touches only assistant `toolCalls` argument values, preserving role/content/tool ids/order/pairing. `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:291`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:319`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:323`\n\n- **Semantic failure points:** semantic throws on missing provider, no compactable prior history, prompt fit failure, empty provider summary, summary fit failure, or final budget overflow. `src/runtime/agent/orchestrator/session/compact/engine.mjs:279`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:315`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:335`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:371`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:384`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:394`\n\n- **Recall failure points:** recall throws on no prior/recall material, empty recall text when not allowed, summary fit failure, or final budget overflow. `src/runtime/agent/orchestrator/session/compact/engine.mjs:653`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:666`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:691`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:697`\n\n- **Loop fallback:** semantic failure can recover via non-LLM unanchored tool-output prune; recall-fasttrack failure is explicitly excluded from that prune path. `src/runtime/agent/orchestrator/session/loop.mjs:501`, `src/runtime/agent/orchestrator/session/loop.mjs:511`, `src/runtime/agent/orchestrator/session/loop.mjs:517`\n\n- **Loop hard fail:** if no compacted result exists after fallback, loop records telemetry/events and throws `agentContextOverflowError`. `src/runtime/agent/orchestrator/session/loop.mjs:552`, `src/runtime/agent/orchestrator/session/loop.mjs:597`, `src/runtime/agent/orchestrator/session/loop.mjs:618`\n\n- **Session fallback:** manual/clear recall-fasttrack failure tries semantic fallback once before recording failure. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:237`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:242`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:245`\n\n- **Session prune/fail:** non-recall manual/clear failures can recover with unanchored prune; otherwise compaction state records `manual_failed`/`post_turn_failed` and returns `changed:false`. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:322`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:348`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:375`\n\n- **State mutation:** successful compaction replaces session messages, clears provider state, records compact metadata, and increments compact count only when changed. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:410`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:411`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:412`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:444`",
-      "card": {
-        "session": "sess_33120_1783058979218_1783058979467_17893d4613374d6fecfb2f6704eff5b2",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 86414,
-        "turns": 7,
-        "tool_calls": 11,
-        "tools_per_turn": 1.6,
-        "total_tool_ms": 1285,
-        "llm_stream_ms": 85250,
-        "cache_ratio": 0.4026380830282444,
-        "cached_tokens": 60928,
-        "prompt_tokens": 151322,
-        "uncached_tokens": 90394,
-        "output_tokens": 5597,
-        "cache_weighted_input_10": 96486.8,
-        "cache_weighted_input_25": 105626,
-        "prompt_growth": 36507,
-        "antipatterns": 0,
-        "issues": 10,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-debug-forwarding-gap#2",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 104643,
-      "sessionId": "sess_47252_1783059063096_1783059063358_bd9cf12a670c04ca4c9e3d75b7938009",
-      "output": "sessionId: sess_47252_1783059063096_1783059063358_bd9cf12a670c04ca4c9e3d75b7938009\n관련 경로를 직접 추적해 활성화부터 첫 전달까지의 손실/지연 지점을 확인하겠습니다.Findings ordered by impact:\n\n1. **Silent drop: first bind to an already-existing transcript can seek to EOF.** `setContext()` defaults `readFileSize/lastFileSize` to current file size unless `replayFromStart` or matching `catchUpFromPersisted` is used (`src/runtime/channels/lib/output-forwarder.mjs:112-129`). On first inbound, `applyTranscriptBinding()` is called without catch-up for an existing transcript (`src/runtime/channels/index.mjs:1597-1603`), so assistant bytes written after `bridge-state.json` but before binding are skipped. **Backlog catch-up: no**, unless a matching persisted cursor already exists or the path goes through the pending-transcript branch.\n\n2. **Delayed, mostly recovered: transcript path known before the `.jsonl` exists.** `startWatch()` cannot watch a missing file and closes silently on `watch.start.catch` (`src/runtime/channels/lib/output-forwarder.mjs:647-670`); `schedulePendingTranscriptRearm()` polls, installs the watcher, then calls `forwardNewText()` when the file appears (`src/runtime/channels/index.mjs:357-390`). **Backlog catch-up: yes within the 60s re-arm window**, because pending binds use start/persisted cursor logic and immediately forward on creation (`src/runtime/channels/index.mjs:469-498`); after timeout, recovery needs another rebind/event.\n\n3. **Delayed: bridge is advertised active before backend ownership is fully connected.** `start()` writes `bridge-state.json` before `refreshBridgeOwnership()`/`backend.connect()` complete (`src/runtime/channels/index.mjs:1734-1747`), while `_isOwner()` rejects forwarding until `bridgeRuntimeConnected && currentOwnerState().owned` (`src/runtime/channels/index.mjs:316`, `src/runtime/channels/lib/output-forwarder.mjs:351-353`). **Backlog catch-up: yes in the normal start path**, because post-connect either re-arms a missing transcript or calls `forwardNewText()` (`src/runtime/channels/index.mjs:801-812`).\n\n4. **Delayed/trigger-dependent: transient ownership-probe false during a watch flush.** `forwardNewText()` returns before reading or advancing the cursor when `_isOwner()` is false (`src/runtime/channels/lib/output-forwarder.mjs:36-42`, `src/runtime/channels/lib/output-forwarder.mjs:351-354`), so bytes are not lost, but the fs.watch edge may be consumed. **Backlog catch-up: partial**, recovered by post-connect catch-up, a later watch event, turn-end/final flush, or rebind; there is no dedicated catch-up when ownership becomes true again while already connected.\n\n5. **Delayed: bind task can run pre-connect and arm a rewatch too early.** Initial binding starts before `backend.connect()` (`src/runtime/channels/index.mjs:758-769`); if the pending re-arm ticks while `bridgeRuntimeConnected` is still false, it exits without rescheduling (`src/runtime/channels/index.mjs:369-375`). **Backlog catch-up: yes in normal startup**, because after connect the code explicitly re-arms the pending path (`src/runtime/channels/index.mjs:803-807`).\n\nVerdict: the main real loss risk is EOF binding of an already-existing transcript without a persisted cursor; the missing-file and pre-connect windows are mostly recovered by re-arm/post-connect catch-up, while ownership false probes can delay until another trigger.",
-      "card": {
-        "session": "sess_47252_1783059063096_1783059063358_bd9cf12a670c04ca4c9e3d75b7938009",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 94112,
-        "turns": 5,
-        "tool_calls": 5,
-        "tools_per_turn": 1,
-        "total_tool_ms": 515,
-        "llm_stream_ms": 93270,
-        "cache_ratio": 0.7144130665653755,
-        "cached_tokens": 92160,
-        "prompt_tokens": 129001,
-        "uncached_tokens": 36841,
-        "output_tokens": 5873,
-        "cache_weighted_input_10": 46057,
-        "cache_weighted_input_25": 59881,
-        "prompt_growth": 35735,
-        "antipatterns": 1,
-        "issues": 9,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "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": 67913,
-      "sessionId": "sess_51160_1783059076138_1783059076364_ce7ebef63b9773c3457e0a8dbeed9dd7",
-      "output": "sessionId: sess_51160_1783059076138_1783059076364_ce7ebef63b9773c3457e0a8dbeed9dd7\n작업 기준 파일과 기존 스크립트 스타일을 먼저 확인하겠습니다.확인한 기준에 맞춰 `package.json`과 새 스크립트를 수정하겠습니다.`package.json`에는 이미 `usage:today` 항목이 있어 새 파일 구현과 검증 중심으로 진행하겠습니다.추가 확인 없이 현재 구조에 맞춰 새 스크립트를 작성하겠습니다.이미 파일이 존재해 내용 확인 후 필요한 차이만 반영하겠습니다.요구사항 대부분이 이미 반영된 상태라, 문법과 실행 검증 후 필요한 보정만 하겠습니다.구현은 이미 있었고, 이제 요구한 검증을 실행하겠습니다.검증은 통과했습니다. 실제 diff를 확인해 요약하겠습니다.재영님, 완료했습니다: `usage:today` 스크립트와 `scripts/usage-today.mjs`가 작업 트리에 반영되어 있습니다.\n\n- **Diff 요약**: `package.json`에 `\"usage:today\": \"node scripts/usage-today.mjs\"` 추가, `scripts/usage-today.mjs` 신규 추가.\n\n- **기능**: `MIXDOG_DATA_DIR`를 우선 사용해 `history/agent-trace.jsonl`을 읽고, 로컬 자정 이후 `usage_raw`/`usage`를 모델별로 집계합니다.\n\n- **출력**: 모델별 `calls`, `input`, `cached`, `output`, `costUsd`와 `GRAND TOTAL` 라인을 출력하며 `--trace`, `--json`도 지원합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 완료. 실행 결과 총합은 `1435 calls`, `53.24M input`, `51.32M cached`, `1.66M output`, `$92.2980`였습니다.",
-      "card": {
-        "session": "sess_51160_1783059076138_1783059076364_ce7ebef63b9773c3457e0a8dbeed9dd7",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 57431,
-        "turns": 10,
-        "tool_calls": 13,
-        "tools_per_turn": 1.3,
-        "total_tool_ms": 2552,
-        "llm_stream_ms": 54609,
-        "cache_ratio": 0.8266361778054949,
-        "cached_tokens": 119296,
-        "prompt_tokens": 144315,
-        "uncached_tokens": 25019,
-        "output_tokens": 3182,
-        "cache_weighted_input_10": 36948.6,
-        "cache_weighted_input_25": 54843,
-        "prompt_growth": 17865,
-        "antipatterns": 1,
-        "issues": 14,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "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": 137375,
-      "sessionId": "sess_58804_1783059088033_1783059088258_fecef461f40816f3c4db38876a81fd3b",
-      "output": " disabled by `compaction.auto=false`, requires a boundary, computes trigger as explicit sub-boundary limit or `boundary-buffer`, and adds request/tool reserve. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `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- **When it fires:** pre-send compaction runs after queued steering is drained, then fires if forced-reactive or `messageTokens + reserve >= trigger`; null estimates force compaction. `src/runtime/agent/orchestrator/session/loop.mjs:323`, `src/runtime/agent/orchestrator/session/loop.mjs:329`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`\n\n- **Budget target:** loop uses `compactTargetBudget()` for compacting, otherwise boundary; target is bounded by boundary and includes reserve. `src/runtime/agent/orchestrator/session/loop.mjs:362`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:163`\n\n- **Reactive path:** provider context-overflow marks one reactive retry, loops back through the normal pre-send compact path, then surfaces overflow if retry was already used. `src/runtime/agent/orchestrator/session/loop.mjs:982`, `src/runtime/agent/orchestrator/session/loop.mjs:995`, `src/runtime/agent/orchestrator/session/loop.mjs:1010`, `src/runtime/agent/orchestrator/session/loop.mjs:1018`\n\n- **Semantic window:** engine sanitizes tool pairs, separates protected/system prefix, finds prior summary, and removes summaries from live history. `src/runtime/agent/orchestrator/session/compact/engine.mjs:100`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:102`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:103`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:110`\n\n- **Semantic tail choice:** with user turns, it preserves up to `tailTurns` recent turns within `preserveRecentBudget`; without user turns, it chooses a non-tool cut point so tool results do not start the tail. `src/runtime/agent/orchestrator/session/compact/engine.mjs:134`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:151`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:240`\n\n- **Semantic output shape:** selected `head` is summarized, `system + summary + tail` is emitted, and mandatory `system+tail` cost can raise the budget rather than fail early. `src/runtime/agent/orchestrator/session/compact/engine.mjs:244`, `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:392`\n\n- **Recall tail choice:** recall-fasttrack keeps the newest two user-anchored turns by default, drops oldest kept turns until under cap, or truncates the surviving turn; no-user tails are valid. `src/runtime/agent/orchestrator/session/compact/engine.mjs:467`, `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:613`\n\n- **Recall pipeline:** loop recall-fasttrack ingests session memory, dumps roots, drains raw rows if needed, caps recall injection to 5% context/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:157`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:161`\n\n- **Redaction hook:** if transcript already fits and not forced, semantic/recall return sanitized messages unchanged; otherwise both redact tool-call secrets before window/tail selection. `src/runtime/agent/orchestrator/session/compact/engine.mjs:286`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:312`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:627`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:644`\n\n- **Redaction semantics:** preserved-tail redaction touches assistant `toolCalls` argument values only, preserves shape/order/pairing, and also redacts raw `key: value` secret strings. `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:219`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:230`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:291`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:319`\n\n- **Prompt-side safety:** summarized head lines use `toolCallSummary()`, whose argument normalization redacts sensitive keys/raw secret strings before truncation. `src/runtime/agent/orchestrator/session/compact/summary.mjs:85`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:121`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:151`\n\n- **Engine failures:** semantic throws for missing provider, empty live/no compactable head, prompt not fitting, empty provider summary, summary not fitting, or final result over budget. `src/runtime/agent/orchestrator/session/compact/engine.mjs:279`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:254`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:315`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:335`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:372`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:385`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:395`\n\n- **Recall failures:** recall-fasttrack throws when there is no compactable head and no recall/prior allowance, recall text is empty, summary cannot fit, or final result exceeds budget. `src/runtime/agent/orchestrator/session/compact/engine.mjs:653`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:666`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:691`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:698`\n\n- **Loop fallback:** non-recall compact errors get one unanchored prune salvage; if that fails, telemetry emits `overflow_failed` and the loop throws `AgentContextOverflowError`. `src/runtime/agent/orchestrator/session/loop.mjs:500`, `src/runtime/agent/orchestrator/session/loop.mjs:517`, `src/runtime/agent/orchestrator/session/loop.mjs:527`, `src/runtime/agent/orchestrator/session/loop.mjs:552`, `src/runtime/agent/orchestrator/session/loop.mjs:618`\n\n- **Manager fallback:** manual/auto-clear runner falls back from recall-fasttrack to semantic, then non-recall unanchored prune, then records `manual_failed`/`post_turn_failed` without replacing messages. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:231`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:242`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:327`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:348`\n\n- **Success effects:** successful loop compaction replaces the in-memory message array, clears provider state, resets loop iteration count, records telemetry, and fires post-compact hook best-effort. `src/runtime/agent/orchestrator/session/loop.mjs:630`, `src/runtime/agent/orchestrator/session/loop.mjs:632`, `src/runtime/agent/orchestrator/session/loop.mjs:639`, `src/runtime/agent/orchestrator/session/loop.mjs:644`, `src/runtime/agent/orchestrator/session/loop.mjs:651`, `src/runtime/agent/orchestrator/session/loop.mjs:718`\n\n- **Crash recovery:** stale persisted `compacting` stage is normalized to `interrupted` on fresh ask/resume so UI/runtime do not treat it as still running. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:36`, `src/runtime/agent/orchestrator/session/manager.mjs:1233`, `src/runtime/agent/orchestrator/session/manager.mjs:1773`",
-      "card": {
-        "session": "sess_58804_1783059088033_1783059088258_fecef461f40816f3c4db38876a81fd3b",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 126887,
-        "turns": 11,
-        "tool_calls": 17,
-        "tools_per_turn": 1.5,
-        "total_tool_ms": 1470,
-        "llm_stream_ms": 125260,
-        "cache_ratio": 0.7053957936861514,
-        "cached_tokens": 273920,
-        "prompt_tokens": 388321,
-        "uncached_tokens": 114401,
-        "output_tokens": 10728,
-        "cache_weighted_input_10": 141793,
-        "cache_weighted_input_25": 182881,
-        "prompt_growth": 74500,
-        "antipatterns": 1,
-        "issues": 17,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "readonly_stall",
-          "broad_tool_result",
-          "service_tier",
-          "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": 89425.8,
-    "turns": 9.3,
-    "tool_calls": 12.5,
-    "tools_per_turn": 1.3,
-    "total_tool_ms": 1626.8,
-    "llm_stream_ms": 87592.7,
-    "cache_ratio": 0.7,
-    "cached_tokens": 154794.7,
-    "prompt_tokens": 210791.7,
-    "uncached_tokens": 55997,
-    "output_tokens": 6391.7,
-    "cache_weighted_input_10": 71476.5,
-    "cache_weighted_input_25": 94695.7,
-    "prompt_growth": 36691.7,
-    "antipatterns": 1.2,
-    "issues": 12.5,
-    "grep_sweep_count": 0
-  },
-  "cards": [
-    {
-      "session": "sess_58408_1783058979193_1783058979442_d47c240b8d093c927461b55a0dc2516c",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 73382,
-      "turns": 5,
-      "tool_calls": 5,
-      "tools_per_turn": 1,
-      "total_tool_ms": 408,
-      "llm_stream_ms": 72483,
-      "cache_ratio": 0.7143255762040293,
-      "cached_tokens": 89600,
-      "prompt_tokens": 125433,
-      "uncached_tokens": 35833,
-      "output_tokens": 6270,
-      "cache_weighted_input_10": 44793,
-      "cache_weighted_input_25": 58233,
-      "prompt_growth": 31998,
-      "antipatterns": 1,
-      "issues": 9,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "broad_tool_result",
-        "service_tier",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_18552_1783058979205_1783058979453_87070f4cdd6037cd39aa2a84fc8356ac",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 98329,
-      "turns": 18,
-      "tool_calls": 24,
-      "tools_per_turn": 1.3,
-      "total_tool_ms": 3531,
-      "llm_stream_ms": 94684,
-      "cache_ratio": 0.8973703724131169,
-      "cached_tokens": 292864,
-      "prompt_tokens": 326358,
-      "uncached_tokens": 33494,
-      "output_tokens": 6700,
-      "cache_weighted_input_10": 62780.4,
-      "cache_weighted_input_25": 106710,
-      "prompt_growth": 23545,
-      "antipatterns": 3,
-      "issues": 16,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "tool_error",
-        "service_tier",
-        "duplicate_tool",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_33120_1783058979218_1783058979467_17893d4613374d6fecfb2f6704eff5b2",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 86414,
-      "turns": 7,
-      "tool_calls": 11,
-      "tools_per_turn": 1.6,
-      "total_tool_ms": 1285,
-      "llm_stream_ms": 85250,
-      "cache_ratio": 0.4026380830282444,
-      "cached_tokens": 60928,
-      "prompt_tokens": 151322,
-      "uncached_tokens": 90394,
-      "output_tokens": 5597,
-      "cache_weighted_input_10": 96486.8,
-      "cache_weighted_input_25": 105626,
-      "prompt_growth": 36507,
-      "antipatterns": 0,
-      "issues": 10,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_47252_1783059063096_1783059063358_bd9cf12a670c04ca4c9e3d75b7938009",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 94112,
-      "turns": 5,
-      "tool_calls": 5,
-      "tools_per_turn": 1,
-      "total_tool_ms": 515,
-      "llm_stream_ms": 93270,
-      "cache_ratio": 0.7144130665653755,
-      "cached_tokens": 92160,
-      "prompt_tokens": 129001,
-      "uncached_tokens": 36841,
-      "output_tokens": 5873,
-      "cache_weighted_input_10": 46057,
-      "cache_weighted_input_25": 59881,
-      "prompt_growth": 35735,
-      "antipatterns": 1,
-      "issues": 9,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "broad_tool_result",
-        "service_tier",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_51160_1783059076138_1783059076364_ce7ebef63b9773c3457e0a8dbeed9dd7",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 57431,
-      "turns": 10,
-      "tool_calls": 13,
-      "tools_per_turn": 1.3,
-      "total_tool_ms": 2552,
-      "llm_stream_ms": 54609,
-      "cache_ratio": 0.8266361778054949,
-      "cached_tokens": 119296,
-      "prompt_tokens": 144315,
-      "uncached_tokens": 25019,
-      "output_tokens": 3182,
-      "cache_weighted_input_10": 36948.6,
-      "cache_weighted_input_25": 54843,
-      "prompt_growth": 17865,
-      "antipatterns": 1,
-      "issues": 14,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_58804_1783059088033_1783059088258_fecef461f40816f3c4db38876a81fd3b",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 126887,
-      "turns": 11,
-      "tool_calls": 17,
-      "tools_per_turn": 1.5,
-      "total_tool_ms": 1470,
-      "llm_stream_ms": 125260,
-      "cache_ratio": 0.7053957936861514,
-      "cached_tokens": 273920,
-      "prompt_tokens": 388321,
-      "uncached_tokens": 114401,
-      "output_tokens": 10728,
-      "cache_weighted_input_10": 141793,
-      "cache_weighted_input_25": 182881,
-      "prompt_growth": 74500,
-      "antipatterns": 1,
-      "issues": 17,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "readonly_stall",
-        "broad_tool_result",
-        "service_tier",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    }
-  ]
-}