thincoder 0.6.0 → 0.7.1

package/src/config.mjs

@@ -5,7 +5,7 @@
  * API key 可用环境变量兜底（未在 providers 中配置时）。
  */
 
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs"
+import { chmodSync, existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs"
 import { homedir } from "node:os"
 import { join } from "node:path"
 
@@ -49,41 +49,45 @@ const DEFAULTS = {
  * 已知模型的能力规格表（前缀匹配，长的在前）。
  * 用于压缩阈值推导、截断续写协议选择、能力感知优化。
  *
- * context:     上下文窗口（tokens）
- * maxOutput:   最大输出 tokens（默认 context）
- * thinking:    是否支持思考/推理模式
- * partialMode: Kimi/Qwen Partial Mode 截断续写（assistant 消息带 partial:true）
- * prefixMode:  DeepSeek Prefix Completion 截断续写（走 /beta 端点，带 prefix:true）
- * multimodal:  是否多模态（支持图片/视觉输入）
- * cacheMode:   上下文缓存方式："auto"=自动/"prompt"=需显式/"none"=不支持
- * thinkApi:    思考模式 API 类型："type"=thinking.type 字段 / "effort"=reasoning_effort 字段
+ * context:           上下文窗口（tokens）
+ * maxOutput:         最大输出 tokens（默认 context）
+ * thinking:          是否支持思考/推理模式
+ * partialMode:       Kimi/Qwen Partial Mode 截断续写（assistant 消息带 partial:true）
+ * prefixMode:        DeepSeek Prefix Completion 截断续写（走 /beta 端点，带 prefix:true）
+ * multimodal:        是否多模态（支持图片/视觉输入）
+ * cacheMode:         上下文缓存方式："auto"=自动/"prompt"=需显式/"none"=不支持
+ * thinkApi:          思考模式 API 类型："type"=thinking.type 字段 / "effort"=reasoning_effort 字段
+ * reasoningEcho:     reasoning_content 跨轮回传策略："required"=必须回传(缺失报错)/"optional"=回传可选(默认不回传)
+ * reasoningEffortEnum: reasoning_effort 合法枚举值（未声明则不校验，原样透传）
+ * tempRange:         temperature 合法范围 [min, max]（未声明则不裁剪）
  */
 const MODEL_SPECS = [
   // DeepSeek V4 系列
-  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "prompt", thinkApi: "type" }],
-  ["deepseek-v4-flash", { context: 256_000,   maxOutput: 384_000, thinking: false, prefixMode: true,  cacheMode: "prompt", thinkApi: "type" }],
-  ["deepseek-reasoner", { context: 256_000,   maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "prompt", thinkApi: "type" }],
-  ["deepseek-chat",     { context: 256_000,   maxOutput: 384_000, thinking: false, prefixMode: true,  cacheMode: "prompt", thinkApi: "type" }],
+  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "prompt", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["high", "max"], tempRange: [0, 2] }],
+  ["deepseek-v4-flash", { context: 256_000,   maxOutput: 384_000, thinking: false, prefixMode: true,  cacheMode: "prompt", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["high", "max"], tempRange: [0, 2] }],
+  ["deepseek-reasoner", { context: 256_000,   maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "prompt", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["high", "max"], tempRange: [0, 2] }],
+  ["deepseek-chat",     { context: 256_000,   maxOutput: 384_000, thinking: false, prefixMode: true,  cacheMode: "prompt", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["high", "max"], tempRange: [0, 2] }],
   // Kimi 系列
-  ["kimi-k3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  partialMode: true, multimodal: true, cacheMode: "prompt", thinkApi: "effort" }],
+  ["kimi-k3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  partialMode: true, multimodal: true, cacheMode: "prompt", thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
   ["kimi-k2",           { context: 256_000,   maxOutput: 128_000, thinking: false, partialMode: true, multimodal: true, cacheMode: "none" }],
   ["moonshot",          { context: 128_000,   maxOutput: 32_000,  thinking: false, cacheMode: "none" }],
   // GLM 系列
-  ["glm-5.2",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type" }],
-  ["glm-5",             { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type" }],
-  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  cacheMode: "auto", thinkApi: "type" }],
+  ["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] }],
+  ["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] }],
+  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1] }],
   // GPT 系列
   ["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" }],
   // Qwen 系列
-  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort" }],
-  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort" }],
-  ["qwen-max",          { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort" }],
-  ["qwen-plus",         { context: 1_000_000, maxOutput: 32_000,  thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort" }],
-  ["qwen",              { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort" }],
+  ["qwen3.8-max-preview", { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["xhigh", "medium", "low"], tempRange: [0, 2] }],
+  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen-max",          { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen-plus",         { context: 1_000_000, maxOutput: 32_000,  thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen",              { context: 1_000_000, maxOutput: 128_000, thinking: false, partialMode: true, cacheMode: "none", thinkApi: "effort", tempRange: [0, 2] }],
   // MiniMax 系列
-  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: false, cacheMode: "auto" }],
-  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: false, cacheMode: "auto" }],
+  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", tempRange: [0, 2] }],
+  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", tempRange: [0, 2] }],
   ["minimax-m1",        { context: 256_000,   maxOutput: 128_000, thinking: false, cacheMode: "auto" }],
 ]
 const DEFAULT_SPEC = { context: 128_000, maxOutput: 32_000, cacheMode: "none" }
@@ -95,7 +99,7 @@ const COMPACT_RATIO = 0.8
 export function specForModel(model) {
   const m = (model ?? "").toLowerCase()
   for (const [prefix, spec] of [...MODEL_SPECS].sort((a,b) => b[0].length - a[0].length)) {
-    if (m.startsWith(prefix)) return spec
+    if (m.startsWith(prefix.toLowerCase())) return spec
   }
   return DEFAULT_SPEC
 }
@@ -111,12 +115,16 @@ export function resolveCompactThreshold(explicit, model) {
 }
 
 /**
- * 从 providers[] 中按 name 查找，找不到返回第一个
+ * 从 providers[] 中按 name 查找。
+ * name 非空但找不到时抛错——activeProvider 打错字静默落到第一个 provider，会拿错 key 打错端点。
+ * name 为空时返回第一个。
  */
 export function findProvider(providers, name) {
   if (name) {
     const found = providers.find((p) => p.name === name)
     if (found) return found
+    const available = providers.map((p) => p.name).join(", ") || "(空)"
+    throw new Error(`activeProvider "${name}" 不在 providers 列表中（可用: ${available}），请检查配置是否打错字: ${configPath}`)
   }
   return providers[0] ?? { name: "default", baseURL: "", model: "" }
 }
@@ -202,5 +210,7 @@ export function loadConfig() {
  */
 export function saveConfig(config) {
   mkdirSync(configDir, { recursive: true })
-  writeFileSync(configPath, JSON.stringify(config, null, 2) + "\n", "utf8")
+  // 0600：config.json 含 API key，不能世界可读（POSIX；Windows 下 chmod 尽力而为）
+  writeFileSync(configPath, JSON.stringify(config, null, 2) + "\n", { encoding: "utf8", mode: 0o600 })
+  try { chmodSync(configPath, 0o600) } catch { /* Windows 上可能失败，忽略 */ }
 }

package/src/context.mjs

@@ -69,13 +69,10 @@ const FALLBACK_NOTE =
 function splitHistory(history) {
   if (history.length <= KEEP_HEAD + KEEP_TAIL + 1) return null
   let headEnd = KEEP_HEAD
-  while (
-    headEnd < history.length &&
-    history[headEnd - 1].role === "assistant" &&
-    history[headEnd - 1].tool_calls?.length &&
-    history[headEnd].role === "tool"
-  ) {
-    headEnd++
+  // head 不能以断头 tool_calls 结尾：assistant 声明了 tool_calls，其 tool 结果必须全部留在 head。
+  // 并行调用时一个 assistant 后面跟多条 tool 消息——只收一条照样 400，必须一次收完
+  if (history[headEnd - 1]?.role === "assistant" && history[headEnd - 1].tool_calls?.length) {
+    while (headEnd < history.length && history[headEnd].role === "tool") headEnd++
   }
   let tailStart = history.length - KEEP_TAIL
 
@@ -158,7 +155,11 @@ export async function compressIfNeeded(agent, threshold) {
   if (tokens <= threshold) return false
 
   const split = splitHistory(history)
-  if (!split) return false
+  if (!split) {
+    // 历史太短（≤13 条）切不出中间段，但 token 已超阈值——典型是一条巨型消息
+    // （大段粘贴/超大注入）。摘要无路可走时退化为确定性瘦身，保证上下文总能减下去
+    return shrinkOversized(agent)
+  }
 
   const middle = history.slice(split.headEnd, split.tailStart)
   const serialized = middle
@@ -189,3 +190,31 @@ export function compressFallback(agent) {
   applyCompression(agent, split.headEnd, split.tailStart, FALLBACK_NOTE)
   return true
 }
+
+/** 单条消息正文的硬截断长度：超过且在压缩无法切分时截断换桩（防一条巨消息卡死压缩） */
+const OVERSIZE_CONTENT_LIMIT = 8_000
+
+/**
+ * 确定性瘦身：splitHistory 切不出中间段（历史太短）但已超阈值时的最后手段，无 LLM 调用。
+ * 把超过 OVERSIZE_CONTENT_LIMIT 的 user/tool 正文截断换桩（保留首尾）；
+ * 不动 reasoning_content（DeepSeek/Kimi 回传协议）与 tool_calls 配对结构，无协议 400 风险。
+ * 只在 compressIfNeeded 判定超阈值后调用。返回是否有消息被截断。
+ */
+export function shrinkOversized(agent) {
+  let shrunk = false
+  for (const m of agent.history) {
+    if ((m.role !== "user" && m.role !== "tool") || typeof m.content !== "string") continue
+    if (m.content.length <= OVERSIZE_CONTENT_LIMIT) continue
+    m.content =
+      m.content.slice(0, 4_000) +
+      `\n[... ${m.content.length - 6_000} chars truncated — single message too large for context window ...]\n` +
+      m.content.slice(-2_000)
+    shrunk = true
+  }
+  if (shrunk) {
+    // 与压缩同理：实测 token 基准随被改动的历史失效，退回估算直到下次响应
+    agent._lastPromptTokens = null
+    agent._usageAtLen = null
+  }
+  return shrunk
+}

package/src/distill.mjs

@@ -67,7 +67,7 @@ export function historyToTranscript(history, { maxChars = 30_000 } = {}) {
     if (m.role === "tool") {
       lines.push(`[工具结果] ${(m.content ?? "").slice(0, 500)}`)
     } else if (m.tool_calls?.length) {
-      const calls = m.tool_calls.map((tc) => `${tc.function.name}(${tc.function.arguments?.slice(0, 200) ?? ""})`).join(", ")
+      const calls = m.tool_calls.map((tc) => `${tc.function?.name ?? "?"}(${tc.function?.arguments?.slice(0, 200) ?? ""})`).join(", ")
       lines.push(`[assistant] ${m.content ?? ""}\n[调用工具] ${calls}`)
     } else {
       lines.push(`[${m.role}] ${m.content ?? ""}`)
@@ -88,7 +88,11 @@ export function historyToTranscript(history, { maxChars = 30_000 } = {}) {
  */
 export async function saveCandidate(memory, candidate, opts = {}) {
   const scope = candidate.scope ?? "personal"
-  const tags = Array.isArray(candidate.tags) ? candidate.tags : (candidate.tags ?? "").split(/\s+/).filter(Boolean)
+  // tags 来自 LLM 输出（不可信）：非数组时先 String 化再按逗号/空白切分——
+  // 直接对非字符串调 .split 会崩，模型也常给 "a, b" 这种逗号串
+  const tags = Array.isArray(candidate.tags)
+    ? candidate.tags.map((t) => String(t)).filter(Boolean)
+    : String(candidate.tags ?? "").split(/[\s,]+/).filter(Boolean)
 
   if (scope === "personal") {
     const id = await put(memory, { type: candidate.type, title: candidate.title, content: candidate.content, tags: tags.join(" ") })

package/src/embedding.mjs

@@ -31,8 +31,15 @@ export async function embed(embedder, texts, { signal } = {}) {
   for (let i = 0; i < texts.length; i += BATCH_SIZE) {
     const batch = texts.slice(i, i + BATCH_SIZE)
     const data = await requestWithRetry(embedder, batch, signal)
-    // API 按 data[].embedding 返回，顺序与输入一致
-    for (const item of data.data) {
+    // 数量不符直接报错——静默接受会让向量与文本错位，污染整个索引
+    if (!Array.isArray(data.data) || data.data.length !== batch.length) {
+      throw new Error(`Embedding API returned ${data.data?.length ?? 0} vectors for ${batch.length} inputs`)
+    }
+    // 规范上 data[] 顺序与输入一致，但以 index 字段为准排序（有的话），不赌服务端实现
+    const items = data.data.every((d) => typeof d.index === "number")
+      ? [...data.data].sort((a, b) => a.index - b.index)
+      : data.data
+    for (const item of items) {
       vectors.push(normalize(Float32Array.from(item.embedding)))
     }
   }
@@ -54,6 +61,8 @@ export function toBlob(vec) {
 
 /** sqlite BLOB → Float32Array */
 export function fromBlob(buf) {
+  // BLOB 可能来自 Buffer 池，byteOffset 不保证 4 对齐，直接建视图会 RangeError——先复制对齐
+  if (buf.byteOffset % 4 !== 0) buf = new Uint8Array(buf)
   return new Float32Array(buf.buffer, buf.byteOffset, buf.byteLength / 4)
 }
 

package/src/gitmem.mjs

@@ -67,7 +67,9 @@ export async function pullTeam(dir) {
  */
 export async function commitAndPush(dir, filename, message) {
   await git(dir, ["add", filename])
-  await git(dir, ["commit", "-m", message])
+  // 内容没变化时 commit 会以 exit 1 报 "nothing to commit"——这是正常的幂等结果，不是错误
+  const dirty = await git(dir, ["status", "--porcelain", "--", filename])
+  if (dirty) await git(dir, ["commit", "-m", message])
   try {
     await git(dir, ["push"])
   } catch {
@@ -80,7 +82,9 @@ export async function commitAndPush(dir, filename, message) {
 async function hasConflict(dir) {
   try {
     const out = await git(dir, ["status", "--porcelain"])
-    return out.split("\n").some((l) => l.startsWith("UU") || l.startsWith("AA") || l.startsWith("DD"))
+    // 未合并状态共 7 种：DD AU UD UA DU AA UU——只看 UU/AA/DD 会漏掉带 U 的四种，
+    // 漏判就不 abort，仓库留在冲突中间态（与"保持仓库干净"的承诺相悖）
+    return out.split("\n").some((l) => l[0] === "U" || l[1] === "U" || l.startsWith("AA") || l.startsWith("DD"))
   } catch {
     return false
   }

package/src/markdown.mjs

@@ -41,14 +41,16 @@ export function parseEntry(text) {
 export function serializeEntry(meta, content) {
   if (!VALID_TYPES.has(meta.type)) throw new Error(`invalid type "${meta.type}"`)
   if (!meta.title) throw new Error("meta.title is required")
-  const tags = (meta.tags ?? []).map((t) => `${t}`).join(", ")
+  // frontmatter 标量必须单行：title/author 含换行会注入伪 frontmatter 行
+  // （如 title "x\ntype: rule" 解析时覆盖真实 type），tags 含换行/逗号同理
+  const tags = (meta.tags ?? []).map((t) => oneLine(t).replaceAll(",", " ")).join(", ")
   const lines = [
     "---",
     `type: ${meta.type}`,
-    `title: ${meta.title}`,
+    `title: ${oneLine(meta.title)}`,
     `tags: [${tags}]`,
-    `author: ${meta.author ?? "unknown"}`,
-    `created: ${meta.created ?? new Date().toISOString().slice(0, 10)}`,
+    `author: ${oneLine(meta.author ?? "unknown")}`,
+    `created: ${oneLine(meta.created ?? new Date().toISOString().slice(0, 10))}`,
   ]
   if (meta.embedding) lines.push(`embedding: ${meta.embedding}`)
   lines.push("---", "", content.trim(), "")
@@ -74,6 +76,11 @@ export function entryFilename(title, date = new Date()) {
 
 // ---------------------------------------------------------------- 内部
 
+/** 压成单行（frontmatter 标量用）：换行折叠为空格，防注入伪字段行 */
+function oneLine(v) {
+  return String(v).replace(/\s*\r?\n\s*/g, " ").trim()
+}
+
 /**
  * 极简 YAML 子集解析：只支持 `key: value` 和 `key: [a, b, c]`。
  * 我们的 frontmatter 是自己生成的，不需要完整 YAML。