@thincoder/core 0.9.4 → 0.9.5

package/model-specs.mjs

@@ -17,7 +17,6 @@
  * partialMode:       Kimi/Qwen Partial Mode truncation continuation (assistant message with partial:true)
  * prefixMode:        DeepSeek Prefix Completion truncation continuation (uses /beta endpoint, with prefix:true)
  * multimodal:        whether multimodal (image/vision input supported)
- * cacheMode:         context caching mode: "auto"=automatic / "prompt"=needs explicit / "none"=unsupported
  * thinkApi:          thinking API type: "type"=thinking.type field / "effort"=reasoning_effort field
  * thinkEnabledValue: when thinkApi is "type", the value used to enable thinking (default "enabled"; MiniMax uses "adaptive")
  * reasoningEcho:     reasoning_content cross-turn echo strategy: "required"=must echo (error if missing) / "optional"=echo optional (default: don't echo)
@@ -30,67 +29,94 @@ const MODEL_SPECS = [
   // beta, automatic disk cache, vision. The two legacy names below are RETIRED — the server
   // still accepts them and serves them from V4.1-Flash today (no switch window), so their
   // rows carry the V4.1-Flash parameters (v4-flash gains multimodal).
-  ["deepseek-flash",    { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-flash",    { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
 // deepseek-v4.1-flash = qwen-plan channel name for DeepSeek V4.1-Flash (token-plan GET /models verified 2026-09-15) — same fields as the row above; ".1" ≠ "-" so plain prefix lookup misses the existing rows (D-PR25)
-  ["deepseek-v4.1-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
-  // deepseek-v4-pro (V4-Pro-0813): fields unchanged — vision capability NOT verified, so no
-  // multimodal (conservative). From 2026-09-14 12:00 Beijing all requests route to V4.1-Flash.
-  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2] }],
+  ["deepseek-v4.1-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  // deepseek-v4-pro (V4-Pro-0813): fields unchanged — 视觉面 **未探**（不声明 multimodal——保守；
+  // 补探登记见 MODEL-SPECS.md §13.4）. From 2026-09-14 12:00 Beijing all requests route to V4.1-Flash.
+  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2] }],
   // deepseek-v4-flash: RETIRED name — still accepted, served by V4.1-Flash today → aligned row
-  ["deepseek-v4-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-v4-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
   // deepseek-v4-flash-vision-exp: retired experimental vision name — still accepted (V4.1-Flash)
-  ["deepseek-v4-flash-vision-exp", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-v4-flash-vision-exp", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
   // Kimi series
-  ["kimi-k3",           { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["kimi-k3",           { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
   // Qwen router prefixes model IDs with provider namespace: kimi/kimi-k3 → kimi-k3 (IK7K4V)
-  ["kimi/kimi-k3",      { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["kimi/kimi-k3",      { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
   // Kimi For Coding endpoint uses the short model ID "k3" (same specs as kimi-k3) — IK5VGJ
-  ["k3",                { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["k3",                { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  // kimi-for-coding / k3-256k / kimi-for-coding-highspeed（Kimi For Coding 端点三名 · 2026-09-25 批 §14.2 #7-9 · 活测 = §1.4 服务端 200 /models）：
+  // context = **活测**逐行；maxOutput = **族沿用 k3 行**（点名来源行——本注即标级）；effort 枚举 ["low","high","max"] = **活测**，
+  // 默认档 max（kimi-for-coding）/ high（k3-256k）落端差表；机制位（partialMode / thinkApi / reasoningEcho）**族沿用 k3 行**
+  // （有则沿、无则不声明）；能力位按 §1.4 现盘口径（image ✓ ⇒ multimodal ✓）。highspeed = 尺寸行 + image ✓（§1.4）——
+  // 活测**无 effort 块** ⇒ thinking / 枚举 / 默认档均不声明；与 kimi-k2.7-code-highspeed（128_000）分值分行（防合并）。
+  ["kimi-for-coding-highspeed", { context: 262_144,   maxOutput: 131_072, multimodal: true }],
+  ["kimi-for-coding",           { context: 1_048_576, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["k3-256k",                   { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  // kimi-k2.6 / kimi-k2.7-code / kimi-k2.7-code-highspeed（三行同形 · 2026-09-24 bench 名单批）：temperature **仅受理 1** = **校验级**
+  // （400 原文「invalid temperature: only 1 is allowed for this model」· roster-expand §2.2）；尺寸位 = **未取证**
+  // （沿现盘兜底值——建行不改变生效值）；**不得声明 tempRange**（KD-31 例外档不变量：入档值 = 实发值）；枚举 / 视觉面 = **未探**。
+  ["kimi-k2.6",                { context: 128_000, maxOutput: 32_000 }],
+  ["kimi-k2.7-code",           { context: 128_000, maxOutput: 32_000 }],
+  ["kimi-k2.7-code-highspeed", { context: 128_000, maxOutput: 32_000 }],
   // GLM series
   // GLM-5.3: thinking always-on (no "disabled"); effort converges to low/high/max — NOT the
-  //          7-level glm-5.2 enum (verified vs docs.bigmodel.cn GLM-5.3 page, 2026-08)
-  ["glm-5.3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5.3-flash",     { context: 1_000_000, maxOutput: 128_000, thinking: true, multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  //          7-level glm-5.2 enum (verified vs docs.bigmodel.cn GLM-5.3 page, 2026-08).
+  //          视觉面 **未探**（不声明 multimodal——保守；补探登记见 MODEL-SPECS.md §13.4）。
+  ["glm-5.3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-5.3-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, multimodal: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
   // glm-5.3-flashx（GLM-5.3-FlashX，2026-09 上线）——独立行（不再蹭 flash 前缀行）：maxOutput 131_072 =
   // **校验级**（400「max_tokens…限制数值范围[1,131072]」）；effort 枚举 ["low","high","max"] = **校验级**
   // （400 原文点名：low/high/max 受理，其余拒绝）。multimodal = **实测**（8×8 纯红 PNG → 答「红色」）。
   // thinking = **实测**且**始终思考**（`thinking:{type:"disabled"}` → 400；裸请求默认开）——off 动作在
   // UI 侧为 no-op（服务端无关闭路径）。其余 = **族沿用** flash 行（context 网络口径未探边；
-  // cacheMode / thinkApi / reasoningEcho / tempRange / noUsageStream = 前缀命中路径日常在用）。
-  // flash 行 128_000 系其自身口径，不得对齐本行（AC-3 零回归）。
-  ["glm-5.3-flashx",    { context: 1_000_000, maxOutput: 131_072, thinking: true, multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5.2",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5",             { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
+  // thinkApi / reasoningEcho / tempRange / noUsageStream = 前缀命中路径日常在用）。
+  // flash 行 maxOutput 131_072 = **校验级**（#20 批 §1.2 同源 · 2026-09-25 批 §14.2 #1 收正）——两行独立（改一行不动另一行）。
+  ["glm-5.3-flashx",    { context: 1_000_000, maxOutput: 131_072, thinking: true, multimodal: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-5.2",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-5",             { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
+  // glm-4.5-air（智谱轻档 · 2026-09-24 bench 名单批）：temperature 0 受理 = **实测**（roster-expand §2.2）；其余 = **族沿用 glm-4 行**
+  // （逐名官方口径未取证）；枚举 / 视觉面 = **未探**（不声明）。medium 直发实测受理（batch-params-judge §1.5）。
+  ["glm-4.5-air", { context: 128_000, maxOutput: 32_000, thinking: true, thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
   // GPT series
-  ["gpt-5.6-sol",       { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "prompt" }],
-  ["gpt-5.6",           { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "prompt" }],
-  ["gpt-4.1",           { context: 1_000_000, maxOutput: 128_000, thinking: false, cacheMode: "prompt" }],
-  ["gpt-4o",            { context: 128_000,   maxOutput: 16_000,  thinking: false, multimodal: true, cacheMode: "prompt" }],
+  ["gpt-5.6-sol",       { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true }],
+  ["gpt-5.6",           { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true }],
+  ["gpt-4.1",           { context: 1_000_000, maxOutput: 128_000, thinking: false }],
+  ["gpt-4o",            { context: 128_000,   maxOutput: 16_000,  thinking: false, multimodal: true }],
   // Qwen series
-  // qwen3.7-max rejects image parts outright (DashScope 400 "Unexpected item type in content") — text-only
-  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["xhigh", "high"], tempRange: [0, 2] }],
+  // qwen-flash / qwen-vl-max（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧实测）：**尺寸行**——maxOutput 32_768 = **实测**
+  // （服务端直报「Range of max_tokens should be [1, 32768]」；131072 / 262144 均 400）。context 未取到 ⇒ 零口径——取兜底值
+  // 128_000（与建行前生效值同值，§14.3 兜底语义续效）；thinking / 视觉有读数（默认关可开 / 不支持）但不落字段（§14.3 五名口径）。
+  ["qwen-flash",        { context: 128_000, maxOutput: 32_768 }],
+  ["qwen-vl-max",       { context: 128_000, maxOutput: 32_768 }],
+  // qwen3.7-max：**实测拒图像**（DashScope 400 "Unexpected item type in content"——真值彩图探针，2026-09-20）
+  // ⇒ 无视觉 = **不声明**（T-7 语义单一：multimodal 仅 true / undefined；补探已毕，无需再探）
+  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true,  thinkApi: "effort", reasoningEffortEnum: ["xhigh", "high"], tempRange: [0, 2] }],
   // qwen3.7-flash / qwen3.8-flash (2026-09-20): thinking + effort enum + maxOutput + image intake
   // are server-MEASURED (batch 2026-09-20-qwen-flash-specs §1.2 — bare request carries reasoning_tokens
   // 166/27; 400 literal lists the enum; max_tokens range [1, 131072]). 3.7-flash enum has NO "max"
   // (6 levels), 3.8-flash adds it (7); enum order = the server's literal order, first member "none"
   // (D-5) — never re-sort by strength. context = doc/web statement (1M class, not API-measured);
-  // partialMode / cacheMode / thinkApi / tempRange = same-family carry-over (not independently
+  // partialMode / thinkApi / tempRange = same-family carry-over (not independently
   // measured on these rows); reasoningEcho stays undeclared (cross-turn echo unverified — R-4).
-  ["qwen3.7-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
-  ["qwen3.8-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
-  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["xhigh", "medium", "low"], tempRange: [0, 2] }],
+  ["qwen3.7-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  // qwen3.7-plus（2026-09-24 bench 名单批）：context = **官方口径（网络转述）**（2026-09-20 qwen-flash-specs 批 AC-6 已裁允许入表——「100 万长文档」族点名）；
+  // maxOutput / partialMode / thinkApi / tempRange / thinking = **族沿用 qwen3.7-flash 行**；枚举 / 视觉面 = **未探**（不声明）。
+  ["qwen3.7-plus",      { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true,  thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen3.8-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["xhigh", "medium", "low"], tempRange: [0, 2] }],
   // qwen3.8-omni-flash / qwen3.8-27b — parent 80-token probe (batch §1.8-①): reasoning_content
   // present, reasoning_tokens 49, `effort:"none"` removes it ⇒ thinking MEASURED; enum / maxOutput
   // / image intake measured the same way. omni-flash: context = doc/web statement (1M class, not
-  // API-measured); partialMode / cacheMode / thinkApi / tempRange = same-family carry-over;
+  // API-measured); partialMode / thinkApi / tempRange = same-family carry-over;
   // reasoningEcho undeclared (cross-turn echo unverified — R-4). The server also takes audio, this
   // repo's call path sends images only ⇒ multimodal below = image intake (no `modalities` field, §2.7).
-  ["qwen3.8-omni-flash",{ context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-omni-flash",{ context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
   // 27b: context 262_144 = doc/web statement, the native value (1M is YaRN extrapolation, local
-  // deploy only); partialMode / cacheMode / thinkApi / tempRange same-family carry-over; reasoningEcho
+  // deploy only); partialMode / thinkApi / tempRange same-family carry-over; reasoningEcho
   // undeclared (cross-turn echo unverified — R-4).
-  ["qwen3.8-27b",       { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-27b",       { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
   // No generic "qwen" fallback row: unlisted qwen names deliberately resolve to DEFAULT_SPEC
   // (128K / 32K, no vision) — per-name disposition in docs/core/design/MODEL-SPECS.md §2.4.
   // Do not re-add a blanket prefix row (silent inheritance is how missing rows stayed invisible).
@@ -100,41 +126,54 @@ const MODEL_SPECS = [
   // ≠ 3.8 系七档）；maxOutput 65_536 = **校验级**（400「Range of max_tokens should be [1, 65536]」；
   // 65536 → 200）。thinking / multimodal = **实测**（裸请求 rc=143–239 默认开、none/minimal rc=0 可关；
   // 64×64 纯红 PNG 五名受理）。context = **官方口径**（未探边，文档口径非 API 实测）；partialMode /
-  // cacheMode / thinkApi / tempRange = **族沿用**（qwen3.7/3.8 行先例）；reasoningEcho 不声明
+  // thinkApi / tempRange = **族沿用**（qwen3.7/3.8 行先例）；reasoningEcho 不声明
   // （跨轮回声未验——R-4）。日期戳快照名不登——前缀命中母名行（有意继承）。
-  ["qwen3.6-flash",     { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-flash",     { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-plus：同形行（共享证据块见 flash 行上）——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位。
-  ["qwen3.6-plus",      { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-plus",      { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-max-preview：同形行——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位；视觉通道实测 ✓ 但纯红图答「黑色」（抽样波动，如实记）。
-  ["qwen3.6-max-preview", { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-max-preview", { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-27b：同形行——校验级 maxOutput / 六档枚举、官方口径 context（未探边）、族沿用机制位。
-  ["qwen3.6-27b",        { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-27b",        { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  // qwen3.5-27b（2026-09-24 bench 名单批）：context 262_144 = **官方口径（定价页分档佐证——[128K, 256K) 档在售；规格页未逐名核）**；
+  // maxOutput / 机制位 = **族沿用 qwen3.6-27b 行**；枚举 / 视觉面 = **未探**（不声明）。
+  ["qwen3.5-27b",        { context: 262_144, maxOutput: 65_536, thinking: true, partialMode: true,  thinkApi: "effort", tempRange: [0, 2] }],
   // qwen3.6-35b-a3b：MoE 后缀（a3b = 激活参数量级）——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位。
-  ["qwen3.6-35b-a3b",    { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-35b-a3b",    { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // MiniMax series
-  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
+  // effort（`reasoning_effort`）在 MiniMax 族 = **无生效面**（**实测**：effort 被忽略——`none` / `medium` 均 rc 在场；
+  // off 路径 = `thinking:{type:"disabled"}` ✓ rc=0）——medium 照发仅口径统一（batch-params-judge §1.2 / KD-32④）。
+  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true,  thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
   // MiMo series (Xiaomi — OpenAI-compatible https://api.xiaomimimo.com/v1; deep thinking via
   // thinking.type, default ON). Family echo policy stays conservative ("required" — tool rounds
   // always echo); 2026-09-22 re-probe: value / missing field / empty string all 200 — the
   // 2026-09-20 "must be passed back" 400 was NOT reproduced. v2.5 rows aligned 2026-09-22:
-  // maxOutput 131_072 = **校验级**; cacheMode "auto" = **实测** (2nd same-prefix round cached 18,816).
-  ["mimo-v2.5-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
-  ["mimo-v2.5",         { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // maxOutput 131_072 = **校验级**.
+  ["mimo-v2.5-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  ["mimo-v2.5",         { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // MiMo V2.6 series (2026-09-22 launch — three independent rows, each probed individually).
   // maxOutput 131_072 / tempRange [0, 1.5] = **校验级** (400 "at most 131072 completion tokens" /
   // "temperature must be within [0, 1.5]", per model); thinking = **实测** (bare request carries
   // reasoning_content; thinking.type disabled → rc gone ⇒ thinkApi "type" = measured face) and
-  // multimodal = **实测** (8×8 pure-red PNG, pro / flash answered "Red"); cacheMode "auto" = **实测**
-  // (cached_tokens 18,688). context 1_000_000 = **官方口径** (docs); reasoningEcho = **族沿用**.
-  ["mimo-v2.6-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // multimodal = **实测** (8×8 pure-red PNG, pro / flash answered "Red"); context 1_000_000 = **官方口径**
+  // (docs); reasoningEcho = **族沿用**.
+  // effort（`reasoning_effort`）在三款 V2.6 = **无生效面**（**实测**：effort 被忽略——`none` / `medium` 均 rc 在场；
+  // off 路径 = `thinking:{type:"disabled"}` ✓ rc=0）——medium 照发仅口径统一（batch-params-judge §1.2 / KD-32④）。
+  ["mimo-v2.6-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // mimo-v2.6-flash: same row shape as pro — maxOutput 131_072 / tempRange [0, 1.5] = **校验级**,
-  // thinking / multimodal / cacheMode = **实测**, context = **官方口径**, reasoningEcho = **族沿用**.
-  ["mimo-v2.6-flash",   { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // thinking / multimodal = **实测**, context = **官方口径**, reasoningEcho = **族沿用**.
+  ["mimo-v2.6-flash",   { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // mimo-v2.6-pro-ultraspeed: same row shape — same grades as flash (**校验级** / **实测** /
   // **官方口径** / **族沿用**); image answer at a 64-token budget was truncated — "Red" at 512.
-  ["mimo-v2.6-pro-ultraspeed", { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
-  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
-  ["minimax-m1",        { context: 256_000,   maxOutput: 128_000, thinking: false, cacheMode: "auto", noUsageStream: true }],
+  ["mimo-v2.6-pro-ultraspeed", { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // minimax-m3（小写前缀行——与上方 MiniMax-M3 行同值，同族两写法）：effort 同按**无生效面**处置
+  // （**实测**：effort 被忽略；off 路径 = `thinking:{type:"disabled"}` ✓ rc=0——batch-params-judge §1.2 / KD-32④）。
+  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true,  thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
+  ["minimax-m1",        { context: 256_000,   maxOutput: 128_000, thinking: false, noUsageStream: true }],
+  // MiniMax-M2.7（前缀行——覆盖标准档与 -highspeed · 2026-09-24 bench 名单批）：thinking = **实测**（裸请求 rc 在场）；off 路径 **未找到**
+  // （`thinking:{type:"disabled"}` 不生效——如实，不声明 thinkApi）；effort 被忽略 ⇒ **无生效面**（枚举不声明；medium 照发仅口径统一）；
+  // 尺寸 / noUsageStream = **族沿用 minimax-m1 行**；视觉面 = **未探**（不声明）。
+  ["MiniMax-M2.7", { context: 256_000, maxOutput: 128_000, thinking: true, noUsageStream: true }],
   // Grok series (xAI — OpenAI-compatible)
   // grok-4.x: 500K context per xAI Grok 4.6 spec (corrected 2026-08; earlier entries said 1M)
   ["grok-4.6",          { context: 500_000,   maxOutput: 64_000,  thinking: false, multimodal: true, tempRange: [0, 2] }],
@@ -145,15 +184,21 @@ const MODEL_SPECS = [
   ["mistral-large",     { context: 128_000,   maxOutput: 32_000,  thinking: false, multimodal: true, tempRange: [0, 2] }],
   ["codestral",         { context: 256_000,   maxOutput: 32_000,  thinking: false, tempRange: [0, 2] }],
   // Claude series (Anthropic)
-  ["claude-opus-5",     { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-sonnet-5",   { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-opus-4",     { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-sonnet-4",   { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-3.5-haiku",  { context: 200_000,   maxOutput: 8_192,   thinking: false, cacheMode: "none", format: "anthropic" }],
+  ["claude-opus-5",     { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-sonnet-5",   { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-opus-4",     { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-sonnet-4",   { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-3.5-haiku",  { context: 200_000,   maxOutput: 8_192,   thinking: false, format: "anthropic" }],
+  // claude-fable-5.1（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 1_000_000 / maxOutput 128_000
+  // （多源网络转述级，本机无该渠道——行值 + 行注即标级）；thinking「always-on adaptive」= 弱源读数 ⇒ 零口径不落字段（§14.3）。
+  ["claude-fable-5.1",  { context: 1_000_000, maxOutput: 128_000 }],
   // Gemini series (Google)
-  ["gemini-3-pro",      { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
-  ["gemini-2.5-pro",    { context: 2_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
-  ["gemini-2.5-flash",  { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
+  ["gemini-3-pro",      { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  ["gemini-2.5-pro",    { context: 2_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  ["gemini-2.5-flash",  { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  // gemini-3.1-pro（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 1_048_576 / maxOutput 65_536（多源
+  // 网络转述级，本机无该渠道——行值 + 行注即标级）；thinking「有·计入输出预算」= 弱源读数 ⇒ 零口径不落字段（§14.3）。
+  ["gemini-3.1-pro",    { context: 1_048_576, maxOutput: 65_536 }],
   // Tencent Hunyuan (TokenHub — 聚合网关 `https://tokenhub.tencentmaas.com/v1`；批 2026-09-20-channel-onboarding)
   // hy3：thinking / thinkApi / effort 枚举 = **本渠道实测**——裸请求 reasoning_content 在场（tok=16），
   // `reasoning_effort:"none"` 即消失（实测唯一有效 off 路径）。枚举 = 七值**受理级探针**（七值全 200 受理；
@@ -163,7 +208,7 @@ const MODEL_SPECS = [
   // **参考实配**（网关对 max_tokens 不硬拒 ⇒ 上限未证）；context 256_000 = **网络口径**（他仓/文档 256K，未本渠道实测）。
   ["hy3",               { context: 256_000, maxOutput: 128_000, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
   // hy3-preview（在场实测：/models + 401 前验活；能力位**未探针**）——**尺寸行**：仅登记尺寸位，取值 =
-  // hy3 **同族沿用**（maxOutput 参考实配 / context 网络口径）。D-11：能力位不跨名沿用 ⇒ thinking / 枚举 / 视觉全不声明。
+  // hy3 **同族沿用**（maxOutput 参考实配 / context 网络口径）。D-11（2026-09-25 收窄现口径）：能力位**无族据**一律不声明 ⇒ 该名无族据 ⇒ thinking / 枚举 / 视觉全不声明。
   ["hy3-preview",       { context: 256_000, maxOutput: 128_000 }],
   // hy4-preview（在场实测：/models + 401；能力位**未探针**）——尺寸行：取值 = hy3 族**同族沿用**
   // （maxOutput 参考实配级取值 / context 网络口径；批次档 §1.2 未给该名尺寸行 ⇒ 上报清单 2，行注即标级）。
@@ -177,8 +222,19 @@ const MODEL_SPECS = [
   // seed-lite = 同族全针（批次档 §1.2 lite 行：思考 tok=161 / 视觉 "Red" / 上限 131 072 / echo 真名）——
   // context 同按 **官方口径** + **实测**下界标注；行独立（同值集 ≠ 同行：改一行不动另一行）。
   ["doubao-seed-2-0-lite-260428", { context: 256_000, maxOutput: 131_072, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], multimodal: true }],
+  // doubao-seed-2-1 三档（火山方舟 · 2026-09-24 bench 名单批 · 三行同形）：thinking / thinkApi / 枚举 = **实测**
+  // （裸请求 rc 在场；`reasoning_effort:"none"` → rc=0；七值全 200、乱值 400 泛化 ⇒ **受理级**——等级低于 2.0 行的
+  // 「真校验」，同值集不混级）；maxOutput 524_288 = **受理级**（262 144 / 524 288 均受理；**上限未证**）；
+  // context = **族沿用 2.0 行**（官方口径未逐名取证）；**视觉面未探**（不声明——如实标级，不冒充实测）。
+  ["doubao-seed-2-1-pro-260915",   { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  ["doubao-seed-2-1-turbo-260628", { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  ["doubao-seed-2-1-lite-260915",  { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  // step-3.7-flash（StepFun · 2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 262_144（多源 256K =
+  // 网络转述级；渠道未激活——400 "product is not activated"）；maxOutput 未取到 ⇒ 零口径——取兜底值 32_000（与建行前
+  // 生效值同值）；行键 = 裸名（渠道名录形 stepfun/step-3.7-flash——命名空间剥离同命中）。
+  ["step-3.7-flash",    { context: 262_144, maxOutput: 32_000 }],
 ]
-const DEFAULT_SPEC = { context: 128_000, maxOutput: 32_000, cacheMode: "none" }
+const DEFAULT_SPEC = { context: 128_000, maxOutput: 32_000 }
 
 /** Look up spec by model name prefix (case-insensitive), conservative default for unknown models.
  *

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@thincoder/core",
-  "version": "0.9.4",
+  "version": "0.9.5",
   "description": "ThinCoder shared core — shared mechanism modules + shared prompt content (CLI ↔ VS Code extension).",
   "keywords": [
     "ai",

package/peer-claims.mjs

@@ -0,0 +1,263 @@
+/**
+ * peer-claims.mjs — R10 多实例协作 L3 **意图认领面**（claims · TTL 租约——MULTI-INSTANCE-COLLAB
+ * §4.4 / D-MI17–D-MI21）。
+ *
+ * 语义（§4.4.1–§4.4.3）：结构化写工具执行成功 ⇒ 本实例登记认领——同 `peers/{sessionId}.json`
+ * 加 `claims` / `claimsUpdatedAt` 两字段，与足迹面 `domains` / `updatedAt` **分字段分存**
+ * （落盘 = 字段级合并写，互不改写，未知字段逐字保留）；租约 `CLAIM_TTL_MS`，同域再写续约
+ * （`claimedAt` 保持首次；覆盖去冗 = 方向性 `covers` 三分支，集内无被覆盖项）；落盘 = 新目标
+ * 即刻 + 续约节流 `CLAIM_RENEW_FLUSH_MS`（届内零 IO）；原子写经 `writeSessionFile`
+ * （.tmp + rename——rename 翻 peers 目录 mtime ⇒ 对端聚合缓存失效，认领回合内可见）。
+ * 释放 = 租约到期 / 属主进程死亡 / 记录被死清理——**无显式释放面**（D-MI19）。
+ *
+ * 读面（§4.4.4）：按 `expiresAt > now` 过滤 + 本实例落盘时剪除（不 unlink——过期不是崩溃
+ * 残留）；属主死亡 ⇒ 整条记录不可见并随既有死清理删除。命中 = 目标 ∩ 他实例未过期认领
+ * （`claimHits`——调用面 = 既有写前预检同点、同一聚合扫描）；软提示文案 `claimNoteText`
+ * （双端逐字一致；`age` / `left` / `who` 规则同 §4.4.4）；去重集 `agent._peerNoted`
+ * （每（目标 × 认领属主）每 run 至多一行——`markClaimNoted` 在提示真正附加时落标记）。
+ *
+ * 本档另归口 peers 目录路径与路径谓词（`pathsOverlap` / `covers`——`peer-domains.mjs`
+ * re-export 保既有 import 面）。测试缝：`_setPeersDirForTest` / `_resetPeersDirForTest`
+ * （沙箱）+ `_setPeerClaimsTestImpl({ nowFn })`（时钟——续约 / 过期两侧零真实等待）。
+ */
+import { existsSync, readFileSync } from "node:fs"
+import { join } from "node:path"
+import { configDir } from "./config.mjs"
+import { normalizeCwd, getSessionId, END, manifestPath, writeSessionFile } from "./session-slots.mjs"
+
+/** 认领租约（§4.4.2）：30 min——同目标再写 / 目标被已有认领覆盖 ⇒ `expiresAt = now + 本值`。 */
+export const CLAIM_TTL_MS = 30 * 60 * 1000
+
+/** 续约落盘节流（§4.4.3）：新目标即刻落盘；纯续约距上次落盘 ≥ 本窗才落盘，其余零 IO。 */
+export const CLAIM_RENEW_FLUSH_MS = 60 * 1000
+
+// ─── peers 目录（测试注入缝——形态先例 = 端侧同名缝）────────────────────────
+let peersDirOverride = null
+export function _setPeersDirForTest(dir) { peersDirOverride = dir }
+export function _resetPeersDirForTest() { peersDirOverride = null }
+
+export function peersDir() {
+  return peersDirOverride ?? join(configDir, "peers")
+}
+
+export function peerFilePath(sessionId) {
+  return join(peersDir(), `${sessionId}.json`)
+}
+
+// ─── 时钟缝（续约 / 过期两侧确定性断言——零真实等待）──────────────────────
+let _testNowFn = null
+
+export function _setPeerClaimsTestImpl({ nowFn = undefined } = {}) {
+  const prev = _testNowFn
+  _testNowFn = nowFn ?? null
+  return prev
+}
+
+export function _resetPeerClaimsTestImpl() {
+  _testNowFn = null
+}
+
+/** 认领面时钟（读面命中判据与落盘同一时钟——时钟缝注入时两侧确定一致；缺省 `Date.now`）。 */
+export function claimsNow() {
+  return _testNowFn ? _testNowFn() : Date.now()
+}
+
+// ─── 路径谓词（自 peer-domains.mjs 归口——单源）────────────────────────────
+
+function normPath(p) {
+  const n = process.platform === "win32" ? String(p).toLowerCase() : String(p)
+  return (process.platform === "win32" ? n.replace(/\//g, "\\") : n.replace(/\\/g, "/")).replace(/[\\/]+$/, "")
+}
+
+/** 路径重叠：a 与 b 相等或互为目录包含（file_ops 目录级操作 / delete 整目录语义——既有
+ *  语义原样：区分大小写按平台；分隔符归一）。本地规范式 = `normPath` 单源。 */
+export function pathsOverlap(a, b) {
+  const pa = normPath(a)
+  const pb = normPath(b)
+  if (pa === pb) return true
+  const sepN = process.platform === "win32" ? "\\" : "/"
+  const prefix = (p, q) => p === q || p.startsWith(q + sepN)
+  return prefix(pa, pb) || prefix(pb, pa)
+}
+
+/** 方向性包含（覆盖去冗判据 §4.4.1）：`coarse` 覆盖 `fine` = 相等或 coarse 为 fine 的目录
+ *  **前缀**（单方向——对称谓词 `pathsOverlap` 不作包含判据）。归一 / 大小写同款。 */
+export function covers(coarse, fine) {
+  const pc = normPath(coarse)
+  const pf = normPath(fine)
+  if (pc === pf) return true
+  return pf.startsWith(pc + (process.platform === "win32" ? "\\" : "/"))
+}
+
+// ─── 认领登记 / 落盘（写成功钩子——§4.4.3）────────────────────────────────
+
+/** 认领状态载体 = `agent._peerClaims`（Map<target, {target, claimedAt, expiresAt}>，与足迹面
+ *  `agent._peerWritten` 同型）；落盘账 = `agent._peerClaimsFlushed`（Set<target>）+
+ *  `agent._peerClaimsFlushedAt`（上次落盘时刻——续约节流锚）。 */
+
+function pendingClaims(agent) {
+  return agent._peerClaims ?? (agent._peerClaims = new Map())
+}
+
+/** 过期剪除（§4.4.2——读面过滤之外的落盘面剪除）：租约已结束的认领不参与覆盖判定，
+ *  也不占据续约对象；剪除同时清落盘账（该目标在盘上已失效 ⇒ 再认领视同新目标）。 */
+function pruneClaims(agent, now) {
+  const set = pendingClaims(agent)
+  const flushed = agent._peerClaimsFlushed
+  for (const [target, claim] of [...set]) {
+    if (claim.expiresAt > now) continue
+    set.delete(target)
+    flushed?.delete(target)
+  }
+}
+
+/**
+ * 认领登记 + 落盘决策（§4.4.3——写工具成功钩子调用；零新手工动作）。三分支（§4.4.1）：
+ * 已有认领 `covers` 新目标 ⇒ 只续约；新目标 `covers` 已有认领项 ⇒ 替换（被覆盖项出集）；
+ * 无包含关系 ⇒ 新目标入集。落盘：含未落盘目标 ⇒ 即刻整写；否则距上次落盘 ≥
+ * `CLAIM_RENEW_FLUSH_MS` ⇒ 续约落盘；其余零 IO。失败容忍（NF2——下次写重试）。
+ */
+export function recordPeerClaims(agent, targets) {
+  try {
+    const now = claimsNow()
+    pruneClaims(agent, now)
+    const set = pendingClaims(agent)
+    for (const target of targets) {
+      const covering = [...set.values()].find((c) => covers(c.target, target))
+      if (covering) {
+        covering.expiresAt = now + CLAIM_TTL_MS // 续约（claimedAt 保持首次）
+        continue
+      }
+      for (const [t, c] of [...set]) {
+        if (covers(target, c.target)) {
+          set.delete(t)
+          agent._peerClaimsFlushed?.delete(t)
+        }
+      }
+      set.set(target, { target, claimedAt: now, expiresAt: now + CLAIM_TTL_MS })
+    }
+    if (set.size === 0) return
+    const flushed = agent._peerClaimsFlushed ?? (agent._peerClaimsFlushed = new Set())
+    const hasUnflushed = [...set.keys()].some((t) => !flushed.has(t))
+    const due = now - (agent._peerClaimsFlushedAt ?? 0) >= CLAIM_RENEW_FLUSH_MS
+    if (hasUnflushed || due) flushPeerClaims(agent, now)
+  } catch {
+    /* 登记失败不影响工具结果（工具主流程永不受认领面影响） */
+  }
+}
+
+/** 读既有实例文件（字段级合并写基线）：缺失 / 损坏 / 结构非法 ⇒ `{}`（按缺失降级——
+ *  损坏件不删，本次落盘覆盖）。 */
+export function readPeerRecord(file) {
+  try {
+    const rec = JSON.parse(readFileSync(file, "utf8"))
+    return rec && typeof rec === "object" && !Array.isArray(rec) ? rec : {}
+  } catch {
+    return {}
+  }
+}
+
+/** 认领落盘（§4.4.1 字段级合并写）：只改 `claims` / `claimsUpdatedAt`——其余字段（含足迹面
+ *  `domains` / `updatedAt`、身份字段、未知字段）逐字保留；身份字段仅在缺失时补齐（本面
+ *  不写他面字段）。原子写 = `writeSessionFile`（.tmp + rename）。落盘门控 = 本 cwd 会话
+ *  manifest 在场（测试卫生——无头 / 未绑定会话不向真实 peers 目录写任何东西；端侧既有
+ *  L3 门同法）。过期条目在落盘时剪除。 */
+export function flushPeerClaims(agent, now = claimsNow()) {
+  try {
+    const cwd = agent?.cwd
+    if (!cwd || !existsSync(manifestPath(cwd))) return
+    const sessionId = getSessionId()
+    const base = readPeerRecord(peerFilePath(sessionId))
+    const claims = [...pendingClaims(agent).values()]
+      .filter((c) => c.expiresAt > now)
+      .map((c) => ({ target: c.target, claimedAt: c.claimedAt, expiresAt: c.expiresAt }))
+      .sort((a, b) => (a.target < b.target ? -1 : a.target > b.target ? 1 : 0))
+    const payload = { ...base }
+    if (payload.sessionId === undefined) payload.sessionId = sessionId
+    if (payload.pid === undefined) payload.pid = process.pid
+    if (payload.end === undefined) payload.end = END
+    if (payload.cwd === undefined) payload.cwd = normalizeCwd(cwd)
+    payload.claims = claims
+    payload.claimsUpdatedAt = now
+    writeSessionFile(peerFilePath(sessionId), payload)
+    agent._peerClaimsFlushed = new Set(claims.map((c) => c.target))
+    agent._peerClaimsFlushedAt = now
+  } catch {
+    /* NF2：失败容忍——下次写重试（落盘账不推进） */
+  }
+}
+
+// ─── 命中判据 / 文案 / 去重（§4.4.4）──────────────────────────────────────
+
+/** 聚合载荷中解析出的认领项形态（`scanPeersDir` 消费——结构非法按缺失）。 */
+export function parseClaims(raw) {
+  if (!Array.isArray(raw)) return []
+  const out = []
+  for (const c of raw) {
+    if (!c || typeof c !== "object") continue
+    if (typeof c.target !== "string" || c.target.length === 0) continue
+    const claimedAt = Number(c.claimedAt) || 0
+    const expiresAt = Number(c.expiresAt) || 0
+    if (expiresAt <= 0) continue
+    out.push({ target: c.target, claimedAt, expiresAt })
+  }
+  return out
+}
+
+/** 命中判据（§4.4.4）：`peers`（聚合载荷——同 cwd / self 已排除 / 属主已判活）中各实例的
+ *  **未过期**认领 ∩ `target` ⇒ `[{end, pid, sessionId, claimedAt, expiresAt}]`（同属主同
+ *  目标至多一条）。纯只读、不抛。 */
+export function claimHits(peers, target, now) {
+  const out = []
+  for (const p of peers ?? []) {
+    const hit = (p.claims ?? []).find((c) => c.expiresAt > now && pathsOverlap(target, c.target))
+    if (hit) out.push({ end: p.end, pid: p.pid, sessionId: p.sessionId, claimedAt: hit.claimedAt, expiresAt: hit.expiresAt })
+  }
+  return out
+}
+
+/** 认领时长（§4.4.4）：`just now`（< 1 min）/ `${n} min ago`。 */
+export function claimAge(claimedAt, now) {
+  const min = Math.floor((now - claimedAt) / 60000)
+  return min < 1 ? "just now" : `${min} min ago`
+}
+
+/** 剩余租约（§4.4.4）：`max(1, ceil((expiresAt − now) / 60000))`（分钟）。 */
+export function claimLeft(expiresAt, now) {
+  return Math.max(1, Math.ceil((expiresAt - now) / 60000))
+}
+
+/** `who` 串（§4.4.4 逐字规则）：认领属主 `${end} pid=${pid}`（无 end 则 `pid=${pid}`）；
+ *  仅足迹命中的属主追加 ` (recent write)`；以 ", " 连接。 */
+export function claimWho(owners) {
+  return owners
+    .map((o) => `${o.end ? `${o.end} ` : ""}pid=${o.pid}${o.recentWrite ? " (recent write)" : ""}`)
+    .join(", ")
+}
+
+/** 认领软提示文案（逐字契约 §4.4.4——双端一致；禁自行解释）。 */
+export function claimNoteText(target, who, age, left) {
+  return `[peer-collab] ${target} — another live instance (${who}) holds a live intent claim on it (claimed ${age}, lease ${left} min left); concurrent edits may overwrite each other. Write not blocked — coordinate before proceeding.`
+}
+
+/** 去重键（§4.4.4）：同一（目标 × 认领属主）每 run 至多一行——键 = `${target}\\u0000${sessionId}`。
+ *  `||`（非 `??`）兜底：缺 `sessionId` 的载荷不得与空串共键（端侧聚合允许空串）。 */
+export function claimNoteKey(target, owner) {
+  return `${target}\u0000${owner.sessionId || owner.pid}`
+}
+
+export function claimNoted(agent, key) {
+  return agent?._peerNoted instanceof Set && agent._peerNoted.has(key)
+}
+
+/** 落去重标记（调用方**在提示真正附加时**调用——写失败路径不消耗额度）。 */
+export function markClaimNoted(agent, keys) {
+  if (!agent || !Array.isArray(keys) || keys.length === 0) return
+  const set = agent._peerNoted instanceof Set ? agent._peerNoted : (agent._peerNoted = new Set())
+  for (const k of keys) set.add(k)
+}
+
+/** 去重集清空（§4.4.4——落点 = `flushPeerDomains` 首步：先于「无写入即返回」早退）。 */
+export function clearClaimNoted(agent) {
+  if (agent?._peerNoted instanceof Set) agent._peerNoted.clear()
+}