dsh-tap 0.20.0

package/scripts/probe-qoder-thinking-config.mjs

@@ -0,0 +1,52 @@
+// probe-qoder-thinking-config.mjs — 取 Qoder 目录里逐模型的思考能力声明
+// （thinking_config）并落证据，同时打印投影出的 reasoningEfforts 档位表。
+//
+// 只读：一次签名 GET（/algo/api/v2/model/list?Encode=1），不发聊天请求、不动配置。
+// 用法：node scripts/probe-qoder-thinking-config.mjs
+// 证据 → docs/probes/qoder-thinking-config-<ts>.json
+import { readJson } from '../core/json-store.js'
+import { homedir } from 'node:os'
+import { dirname, join } from 'node:path'
+import { fileURLToPath } from 'node:url'
+import { mkdirSync, writeFileSync } from 'node:fs'
+
+const ROOT = join(dirname(fileURLToPath(import.meta.url)), '..')
+const store = readJson(join(homedir(), '.dsh', 'qoder-plugin-auth.json'))
+const { createCosyRuntime } = await import('../providers/qoder/cosy.js')
+const { fetchQoderCatalog, qoderReasoningEfforts } = await import('../providers/qoder/catalog.js')
+
+const cosy = createCosyRuntime({ wasmPath: join(ROOT, 'providers', 'qoder', 'qoder_auth.wasm') })
+const cred = { accessToken: store.auth.accessToken, machineId: store.machine.machineId, uid: store.account.uid }
+const infer = process.argv[2] ?? 'https://gateway.qoder.com.cn'
+const cat = await fetchQoderCatalog(cosy, cred, infer)
+
+const entries = {}
+for (const [id, entry] of Object.entries(cat.entries)) {
+  entries[id] = {
+    is_reasoning: entry.is_reasoning === true,
+    thinking_config: entry.thinking_config ?? null,
+    effortTable: qoderReasoningEfforts(entry),
+  }
+}
+
+console.log('model                     is_reasoning  effortTable')
+for (const [id, e] of Object.entries(entries)) {
+  const table = e.effortTable
+  const shape = table === null
+    ? '—'
+    : Object.entries(table).map(([k, v]) => `${k}${v === null ? ':null' : `:${v}`}`).join('/')
+  console.log(`${id.padEnd(24)} ${String(e.is_reasoning).padEnd(13)} ${shape}`)
+}
+const withTable = Object.values(entries).filter((e) => e.effortTable !== null).length
+console.log(`\n${withTable}/${Object.keys(entries).length} 个模型出档位表（宿主 Model/Effort 选择器据此出档）`)
+
+mkdirSync(join(ROOT, 'docs', 'probes'), { recursive: true })
+const out = join(ROOT, 'docs', 'probes', `qoder-thinking-config-${Date.now()}.json`)
+writeFileSync(out, JSON.stringify({
+  at: new Date().toISOString(),
+  source: `GET ${infer}/algo/api/v2/model/list?Encode=1（providers/qoder/catalog.js，签名只读）`,
+  models: Object.keys(entries).length,
+  withEffortTable: withTable,
+  entries,
+}, null, 2))
+console.log('证据 →', out)

package/scripts/probe-qoder-thinking-efforts.mjs

@@ -0,0 +1,269 @@
+#!/usr/bin/env node
+/**
+ * probe-qoder-thinking-efforts.mjs — Qoder 本地翻译网关（OpenAI 兼容 /v1）
+ * reasoning_effort 档位单调性实测：显式 effort 各档 × N 采样，聚合 reasoning
+ * 字符数 / completion tokens，按档位语义序打单调性判定。
+ *
+ * 为什么需要：0.18.0/0.19.0 记录在案的未定论——6 臂单采样非单调、disabled 臂仍
+ * 出 619 字推理 ⇒ 按踩坑 #42（能力声明 ≠ 线值）没敢下结论。本探针把那次取证
+ * 升级为「显式档位 × 多采样聚合」，一条命令出 MONOTONIC / INCONCLUSIVE。
+ *
+ * 臂设计（变量干净优先）：
+ *   - 臂 = **显式 reasoning_effort**。网关对客户端已带的 effort 透传不覆盖
+ *     （providers/qoder/gateway.js：仅 payload 未带时才注入该模型 prefs 的
+ *     effort），显式带参即绕开 prefs 注入，baseline 不被存量 max 污染。
+ *   - off 臂跳过：openai 方言 off = 省略参数，省略会触发网关注入 prefs effort，
+ *     变量不干净。off 档是否真能关思考不在本探针范围（历史 disabled 臂实测
+ *     仍出推理），报告按「off 未测」处理。
+ *   - 各模型跑哪些档 = 目录声明快照（DECLARED_TIERS，真源 providers/qoder/
+ *     catalog.js qoderReasoningEfforts；2026-10-04 全量目录实测，见
+ *     docs/probes/qoder-thinking-config-*.json，可用
+ *     `node scripts/probe-qoder-thinking-config.mjs` 重取）减 off，按语义序
+ *     off<low<medium<high<xhigh<max 排序后做相邻对单调性检查。
+ *
+ * 判据（踩坑 #42 纪律：多采样、非单调不下结论）：
+ *   每臂 --repeat（默认 3）次同题同 prompt → 聚合 reasoningChars min/mean/max
+ *   与 completionTokens min/mean/max；按档位语义序检查**均值单调不减**——
+ *     任一相邻对下降   → INCONCLUSIVE（#42：非单调，不下结论）
+ *     任一臂有样本报错 → INCONCLUSIVE（错误原样落证据）
+ *     全部不减        → MONOTONIC（若各档均值全等，额外标注 flat）
+ *
+ * 用法：
+ *   node scripts/probe-qoder-thinking-efforts.mjs                  # 默认 qmodel_38max+dmodel ×3 repeat（15 次）
+ *   node scripts/probe-qoder-thinking-efforts.mjs --models dmodel  # 只跑一个模型
+ *   node scripts/probe-qoder-thinking-efforts.mjs --repeat 5       # 每臂 5 次采样
+ *   node scripts/probe-qoder-thinking-efforts.mjs --endpoint http://127.0.0.1:3903/v1
+ *   node scripts/probe-qoder-thinking-efforts.mjs --help
+ *
+ * 额度账：总调用 = Σ(每模型档位臂数) × --repeat；默认 2 模型（3+2 臂）× 3 = 15
+ * 次，硬闸 ≤ 30 次超出即拒跑。max_tokens 钳 1024。走的是 Qoder 订阅额度
+ * （免费输入面），每臂样本真实入账。
+ *
+ * 凭据：无需配置——鉴权走本地网关固定 token（Authorization: Bearer dsh-qoder-bridge，
+ * 网关仅监听 127.0.0.1 回环），Qoder 侧凭据由桌面端 OAuth 会话提供。
+ * 证据 → docs/probes/qoder-efforts-<ts>.json（含 endpoint/模型/臂/逐样本原始 usage）。
+ */
+import { mkdirSync, writeFileSync } from 'node:fs'
+import { dirname, join } from 'node:path'
+import { fileURLToPath } from 'node:url'
+
+const ROOT = join(dirname(fileURLToPath(import.meta.url)), '..')
+const ENDPOINT_DFLT = process.env.QODER_BRIDGE_ENDPOINT ?? 'http://127.0.0.1:3913/v1'
+const BRIDGE_TOKEN = 'dsh-qoder-bridge'
+const MAX_TOKENS = 1024
+const MAX_TOTAL_CALLS = 30
+const CALL_TIMEOUT_MS = 180_000
+const PROMPT = '一个农夫有 17 只羊，除了 9 只以外都跑了，还剩几只？只给最终数字。'
+
+/** 目录声明档位静态快照（DECLARED_TIERS，来源见头注释）。目录改版后重取。 */
+const DECLARED_TIERS = {
+  qmodel_38max: ['off', 'low', 'medium', 'xhigh'],
+  qfmodel: ['off', 'low', 'medium', 'xhigh'],
+  dmodel: ['off', 'high', 'max'],
+  gm51model: ['off', 'high', 'max'],
+  gmodel: ['high', 'low', 'max'],
+  kmodel: ['high', 'low', 'max'],
+  kmodel_latest: ['high', 'low', 'max'],
+}
+/** 档位语义序（只用于排序做单调性检查；各家族内部顺序均成立）。 */
+const TIER_RANK = { off: 0, low: 1, medium: 2, high: 3, xhigh: 4, max: 5 }
+
+const args = process.argv.slice(2)
+const argOf = (name, dflt) => {
+  const i = args.indexOf(name)
+  return i >= 0 && args[i + 1] !== undefined && !args[i + 1].startsWith('--') ? args[i + 1] : dflt
+}
+const ENDPOINT = String(argOf('--endpoint', ENDPOINT_DFLT)).replace(/\/+$/, '')
+const MODELS = String(argOf('--models', 'qmodel_38max,dmodel')).split(',').map((s) => s.trim()).filter(Boolean)
+const REPEAT = Number(argOf('--repeat', '3'))
+
+const USAGE = `probe-qoder-thinking-efforts.mjs — Qoder 本地网关 reasoning_effort 档位单调性实测
+
+用法：node scripts/probe-qoder-thinking-efforts.mjs [选项]
+
+臂 = 显式 reasoning_effort（透传不覆盖，绕开 prefs 注入）；off 臂跳过（openai 方言
+off = 省略参数，会触发 prefs 注入，变量不干净）。各模型档位取目录声明快照减 off。
+判据：每臂多采样聚合，档位语义序上检查 reasoning 字符数均值单调不减——
+非单调 = INCONCLUSIVE（踩坑 #42）；单调 = MONOTONIC。
+
+  --models a,b       指定模型（默认 qmodel_38max,dmodel；可选 qfmodel/gm51model/gmodel/kmodel/kmodel_latest）
+  --repeat N         每臂采样 N 次并聚合 min/mean/max（默认 3）
+  --endpoint URL     本地 Qoder 网关（默认 http://127.0.0.1:3913/v1，或环境变量 QODER_BRIDGE_ENDPOINT）
+  --help             本帮助
+
+总额度硬闸 30 次上游调用；max_tokens 钳 1024；同题同 prompt 顺序采样。
+证据 → docs/probes/qoder-efforts-<ts>.json。`
+
+if (args.includes('--help') || args.includes('-h')) {
+  console.log(USAGE)
+  process.exit(0)
+}
+
+if (!Number.isInteger(REPEAT) || REPEAT < 1) {
+  console.error(`--repeat 必须是 ≥1 的整数，收到：${argOf('--repeat', '3')}`)
+  process.exit(2)
+}
+
+// 每模型的臂 = 目录声明减 off，按语义序排（含 off 时注明跳过原因）。
+function armsOf(model) {
+  const tiers = DECLARED_TIERS[model]
+  if (!tiers) return { error: `DECLARED_TIERS 无该模型的声明快照（已知：${Object.keys(DECLARED_TIERS).join(', ')}）` }
+  const run = tiers.filter((t) => t !== 'off')
+  const unsortable = run.filter((t) => !(t in TIER_RANK))
+  if (unsortable.length) return { error: `档位拼写无法排序：${unsortable.join(',')}（TIER_RANK 需扩充）` }
+  run.sort((a, b) => TIER_RANK[a] - TIER_RANK[b])
+  return { arms: run, skippedOff: tiers.includes('off') }
+}
+
+const plan = MODELS.map((model) => ({ model, ...armsOf(model) }))
+const badPlan = plan.filter((p) => p.error)
+if (badPlan.length) {
+  for (const p of badPlan) console.error(`${p.model}: ${p.error}`)
+  process.exit(2)
+}
+const totalCalls = plan.reduce((n, p) => n + p.arms.length * REPEAT, 0)
+console.log('额度账：')
+for (const p of plan) console.log(`  ${p.model}: ${p.arms.join(' < ')} 共 ${p.arms.length} 臂 × ${REPEAT} repeat = ${p.arms.length * REPEAT} 次`)
+console.log(`合计 ${totalCalls} 次（硬闸 ${MAX_TOTAL_CALLS}）${plan.some((p) => p.skippedOff) ? '；off 臂跳过（省略参数会触发 prefs 注入，变量不干净）' : ''}`)
+if (totalCalls > MAX_TOTAL_CALLS) {
+  console.error(`超出硬闸：${totalCalls} > ${MAX_TOTAL_CALLS} 次上游调用，拒跑。请减模型/臂或调低 --repeat。`)
+  process.exit(2)
+}
+
+// 预检：本地 GET /v1/models 免费且不烧上游额度（网关本地路由），验证可达 + 目录在位。
+async function preflight() {
+  let res
+  try {
+    res = await fetch(`${ENDPOINT}/models`, { headers: { authorization: `Bearer ${BRIDGE_TOKEN}` }, signal: AbortSignal.timeout(10_000) })
+  } catch (err) {
+    throw new Error(`网关不可达 ${ENDPOINT}/models：${err?.cause?.code ?? err?.message ?? err}`)
+  }
+  const text = await res.text()
+  let json = null
+  try { json = JSON.parse(text) } catch { /* keep text */ }
+  if (!res.ok || !Array.isArray(json?.data)) {
+    throw new Error(`预检失败 HTTP ${res.status}：${text.slice(0, 300)}`)
+  }
+  return json.data.map((m) => m?.id).filter(Boolean)
+}
+
+let catalogIds
+try {
+  catalogIds = await preflight()
+} catch (err) {
+  console.error(String(err?.message ?? err))
+  process.exit(2)
+}
+console.log(`预检 OK：${ENDPOINT}（目录 ${catalogIds.length} 个模型）`)
+const missing = plan.filter((p) => !catalogIds.includes(p.model))
+if (missing.length) {
+  console.error(`模型不在网关目录里：${missing.map((p) => p.model).join(', ')}（目录有：${catalogIds.join(', ')}）`)
+  process.exit(2)
+}
+
+async function call(model, effort) {
+  let res
+  try {
+    res = await fetch(`${ENDPOINT}/chat/completions`, {
+      method: 'POST',
+      headers: { authorization: `Bearer ${BRIDGE_TOKEN}`, 'content-type': 'application/json' },
+      body: JSON.stringify({
+        model,
+        max_tokens: MAX_TOKENS,
+        reasoning_effort: effort,
+        messages: [{ role: 'user', content: PROMPT }],
+      }),
+      signal: AbortSignal.timeout(CALL_TIMEOUT_MS),
+    })
+  } catch (err) {
+    return { status: 0, reasoningChars: 0, textChars: 0, completionTokens: null, usage: null, error: { code: 'fetch_failed', message: String(err?.cause?.code ?? err?.message ?? err) } }
+  }
+  const text = await res.text()
+  let json = null
+  try { json = JSON.parse(text) } catch { /* keep text */ }
+  const msg = json?.choices?.[0]?.message ?? null
+  return {
+    status: res.status,
+    reasoningChars: typeof msg?.reasoning_content === 'string' ? msg.reasoning_content.length : 0,
+    textChars: typeof msg?.content === 'string' ? msg.content.length : 0,
+    completionTokens: Number.isFinite(json?.usage?.completion_tokens) ? json.usage.completion_tokens : null,
+    finishReason: json?.choices?.[0]?.finish_reason ?? null,
+    usage: json?.usage ?? null,
+    error: json?.error
+      ? { code: json.error.code ?? null, message: String(json.error.message ?? '').slice(0, 200) }
+      : (res.ok ? null : { code: `http_${res.status}`, message: text.slice(0, 200) }),
+  }
+}
+
+// 每臂聚合：保留全部逐样本原始数据，标量归并（min/mean/max，同 probe-ark-thinking
+// 的 aggregateArm 思路）。completionTokens 逐样本可能为 null，只对数值聚合。
+function aggregateArm(samples) {
+  const stat = (pick) => {
+    const vs = samples.map(pick).filter((v) => Number.isFinite(v))
+    if (!vs.length) return null
+    return { min: Math.min(...vs), mean: Math.round((vs.reduce((a, b) => a + b, 0) / vs.length) * 10) / 10, max: Math.max(...vs) }
+  }
+  const statuses = [...new Set(samples.map((s) => s.status))]
+  const errors = [...new Set(samples.filter((s) => s.error).map((s) => JSON.stringify(s.error)))].map((s) => JSON.parse(s))
+  return {
+    runs: samples.length,
+    status: statuses.length === 1 ? statuses[0] : statuses,
+    reasoningChars: stat((s) => s.reasoningChars),
+    completionTokens: stat((s) => s.completionTokens),
+    error: errors.length ? errors : null,
+    samples,
+  }
+}
+
+const fmt = (a) => `http=${JSON.stringify(a.status)} ×${a.runs} reasoning=min ${a.reasoningChars?.min ?? '—'}/mean ${a.reasoningChars?.mean ?? '—'}/max ${a.reasoningChars?.max ?? '—'} completion=min ${a.completionTokens?.min ?? '—'}/mean ${a.completionTokens?.mean ?? '—'}/max ${a.completionTokens?.max ?? '—'}${a.error ? ` error=${JSON.stringify(a.error)}` : ''}`
+
+// 单调性判定（#42）：先看臂级错误，再看档位语义序上均值是否单调不减。
+function judge(armNames, arms) {
+  const bad = armNames.filter((a) => arms[a].error)
+  if (bad.length) return { verdict: 'INCONCLUSIVE', reason: `臂报错：${bad.join(',')}（错误原文见证据 samples）` }
+  const means = armNames.map((a) => arms[a].reasoningChars?.mean)
+  if (means.some((v) => !Number.isFinite(v))) return { verdict: 'INCONCLUSIVE', reason: 'usage/reasoning 缺失，无法聚合' }
+  for (let i = 1; i < means.length; i++) {
+    if (means[i] < means[i - 1]) return { verdict: 'INCONCLUSIVE', reason: `非单调（#42）：${armNames[i - 1]} mean ${means[i - 1]} → ${armNames[i]} mean ${means[i]} 下降，不下结论` }
+  }
+  const flat = new Set(means).size === 1
+  return { verdict: 'MONOTONIC', reason: flat ? '均值随档位单调不减（全平：各档观测相等，未见档位差异）' : '均值随档位单调不减' }
+}
+
+const report = []
+for (const { model, arms: armNames, skippedOff } of plan) {
+  const arms = {}
+  for (const arm of armNames) {
+    const samples = []
+    for (let i = 0; i < REPEAT; i++) samples.push(await call(model, arm))
+    arms[arm] = aggregateArm(samples)
+  }
+  const { verdict, reason } = judge(armNames, arms)
+  report.push({
+    model,
+    declaredTiers: DECLARED_TIERS[model],
+    armsRun: armNames,
+    offSkipped: skippedOff || undefined,
+    verdict,
+    verdictReason: reason,
+    arms,
+  })
+  console.log(`\n### ${model} → ${verdict}（${reason}）`)
+  for (const arm of armNames) console.log(`  ${arm.padEnd(7)}: ${fmt(arms[arm])}`)
+}
+
+mkdirSync(join(ROOT, 'docs', 'probes'), { recursive: true })
+const out = join(ROOT, 'docs', 'probes', `qoder-efforts-${Date.now()}.json`)
+writeFileSync(out, JSON.stringify({
+  at: new Date().toISOString(),
+  endpoint: ENDPOINT,
+  purpose: 'reasoning_effort 档位单调性实测（0.18.0/0.19.0 未定论的复核：显式档位 × 多采样）',
+  prompt: PROMPT,
+  maxTokens: MAX_TOKENS,
+  repeat: REPEAT,
+  armDesign: '臂 = 显式 reasoning_effort（网关透传不覆盖，绕开 prefs 注入）；off 臂跳过（省略参数会触发 prefs 注入）',
+  declaredTiersSource: 'providers/qoder/catalog.js qoderReasoningEfforts 的 2026-10-04 目录快照（docs/probes/qoder-thinking-config-*.json）',
+  verdictRule: '档位语义序（off<low<medium<high<xhigh<max）上 reasoningChars 均值单调不减 → MONOTONIC；任一下降或臂报错 → INCONCLUSIVE（踩坑 #42）',
+  models: report,
+}, null, 2))
+console.log('\n证据 →', out)

package/scripts/probe-quota.mjs

@@ -0,0 +1,136 @@
+#!/usr/bin/env node
+/**
+ * Quota-signal probe: what can the plugin's credentials actually learn about
+ * the account's remaining quota? Probes every reachable route and records
+ * the evidence (docs/probes/). Non-chat probes are free; the chat-header
+ * check costs one minimal completion (max_tokens=1, deepseek-v3 ≈0.001 credit).
+ *
+ *   GET  /v2/accounts                        → account metadata (plan type…)
+ *   POST /v2/billing/meter/get-dosage-notify → low-quota banner (the official
+ *                                              CLI's BillingService source)
+ *   POST /v2/chat/completions (max_tokens=1) → response headers scanned for
+ *                                              quota/rate-limit fields
+ *
+ * Background (2026-08-18): quota is account-level and shared with WorkBuddy
+ * (same Tencent-Cloud.coding-copilot auth realm; workbuddy.cn serves the
+ * isomorphic device flow at /v2/plugin/auth/state). The web console's numeric
+ * plan API (codebuddy.cn/profile/plan) is cookie-authed — not reachable with
+ * CLI OAuth tokens or ck_ keys.
+ *
+ * Usage: node scripts/probe-quota.mjs [--out docs/probes/quota-YYYY-MM-DD.json]
+ *        [--model deepseek-v3] [--skip-chat]
+ * Credentials resolve exactly like the plugin's: active key from
+ * ~/.dsh/codebuddy-plugin.json (fallback: CODEBUDDY_API_KEY / credentials file);
+ * in OAuth mode the stored access token is used.
+ */
+
+import { readFileSync, existsSync, writeFileSync, mkdirSync, appendFileSync } from 'node:fs'
+import { homedir } from 'node:os'
+import { join, dirname } from 'node:path'
+
+const DSH_HOME = process.env.DSH_HOME ?? join(homedir(), '.dsh')
+const GATEWAY = 'https://copilot.tencent.com'
+const UA = 'CLI/unknown CodeBuddy/2.136.0'
+const argValue = (flag) => {
+  const i = process.argv.indexOf(flag)
+  return i > 0 ? process.argv[i + 1] : null
+}
+const MODEL = argValue('--model') ?? 'deepseek-v3'
+const SKIP_CHAT = process.argv.includes('--skip-chat')
+
+function resolveCredential() {
+  try {
+    const cfg = JSON.parse(readFileSync(join(DSH_HOME, 'codebuddy-plugin.json'), 'utf8'))
+    if (cfg.authMode === 'oauth') {
+      const auth = JSON.parse(readFileSync(join(DSH_HOME, 'codebuddy-plugin-auth.json'), 'utf8'))
+      if (auth?.auth?.accessToken) return { authorization: `Bearer ${auth.auth.accessToken}`, apiKey: null }
+    }
+    const active = (cfg.apiKeys ?? []).find((k) => k.name === cfg.activeApiKey)
+    if (active?.key) return { authorization: `Bearer ${active.key}`, apiKey: active.key }
+  } catch { /* fall through */ }
+  if (process.env.CODEBUDDY_API_KEY) {
+    return { authorization: `Bearer ${process.env.CODEBUDDY_API_KEY}`, apiKey: process.env.CODEBUDDY_API_KEY }
+  }
+  const credFile = join(DSH_HOME, '.credentials.yaml')
+  if (existsSync(credFile)) {
+    const m = readFileSync(credFile, 'utf8').match(/^\s*CODEBUDDY_API_KEY:\s*["']?([^"'\s]+)["']?\s*$/m)
+    if (m) return { authorization: `Bearer ${m[1]}`, apiKey: m[1] }
+  }
+  throw new Error('no CodeBuddy credential found')
+}
+
+async function probe(name, { method = 'GET', path, body = null, headers = {} }) {
+  const t0 = Date.now()
+  try {
+    const res = await fetch(`${GATEWAY}${path}`, {
+      method,
+      headers: { Accept: 'application/json', 'User-Agent': UA, ...headers },
+      body: body ? JSON.stringify(body) : undefined,
+    })
+    const text = await res.text()
+    let json = null
+    try { json = JSON.parse(text) } catch { /* HTML error page */ }
+    return { name, method, path, status: res.status, ms: Date.now() - t0, body: json ?? text.slice(0, 200) }
+  } catch (err) {
+    return { name, method, path, ms: Date.now() - t0, error: String(err?.message ?? err) }
+  }
+}
+
+const cred = resolveCredential()
+const authHeaders = { Authorization: cred.authorization, ...(cred.apiKey ? { 'x-api-key': cred.apiKey } : {}) }
+
+const out = { at: new Date().toISOString(), gateway: GATEWAY, results: [] }
+
+console.log('quota probe: /v2/accounts + get-dosage-notify' + (SKIP_CHAT ? '' : ` + chat headers (${MODEL})`))
+
+out.results.push(await probe('accounts', { path: '/v2/accounts', headers: authHeaders }))
+out.results.push(await probe('dosage-notify', {
+  method: 'POST',
+  path: '/v2/billing/meter/get-dosage-notify',
+  headers: { ...authHeaders, 'Content-Type': 'application/json' },
+  body: {},
+}))
+
+if (!SKIP_CHAT) {
+  // Minimal streamed chat: capture the RESPONSE HEADERS only and look for
+  // quota/rate-limit fields (the usage body is covered by the bridge meter).
+  const t0 = Date.now()
+  const res = await fetch(`${GATEWAY}/v2/chat/completions`, {
+    method: 'POST',
+    headers: { ...authHeaders, 'Content-Type': 'application/json', 'User-Agent': UA },
+    body: JSON.stringify({
+      model: MODEL, stream: true, stream_options: { include_usage: true }, max_tokens: 1,
+      messages: [{ role: 'user', content: 'hi' }],
+    }),
+  })
+  const headers = {}
+  res.headers.forEach((v, k) => { headers[k] = v })
+  await res.text()
+  const quotaish = Object.keys(headers).filter((k) => /quota|credit|limit|remain|balance|dosage/i.test(k))
+  out.results.push({
+    name: 'chat-response-headers',
+    method: 'POST',
+    path: '/v2/chat/completions',
+    status: res.status,
+    ms: Date.now() - t0,
+    quotaRelatedHeaders: quotaish,
+    quotaRelatedHeaderCount: quotaish.length,
+    headers,
+  })
+}
+
+for (const r of out.results) {
+  const summary = r.error
+    ? `error=${r.error}`
+    : `http=${r.status} ${JSON.stringify(r.quotaRelatedHeaderCount !== undefined ? { quotaHeaders: r.quotaRelatedHeaderCount } : r.body).slice(0, 300)}`
+  console.log(`${r.name.padEnd(22)} ${summary}`)
+}
+
+const outPath = argValue('--out')
+if (outPath) {
+  mkdirSync(dirname(outPath), { recursive: true })
+  if (existsSync(outPath)) appendFileSync(outPath, '\n')
+  writeFileSync(outPath, JSON.stringify(out, null, 2) + '\n', { flag: 'a' })
+  console.log(`evidence appended → ${outPath}`)
+}
+console.log('probe done.')

package/scripts/probe-quota2.mjs

@@ -0,0 +1,144 @@
+#!/usr/bin/env node
+/**
+ * Quota-signal map probe (topic 4 of docs/rules/): WHICH endpoints and WHICH
+ * response headers carry quota information?
+ *
+ * Prior inventory (AGENTS.md:26, docs/probes/quota-2026-08-18.json):
+ *   - GET /v2/accounts → account metadata (plan type) in body
+ *   - POST /v2/billing/meter/get-dosage-notify → low-quota banner source;
+ *     healthy shape {dosageNotifyCode:0, dosageNotifyZh:"", ...}
+ *   - chat response headers: 12 enumerated, ZERO quota-ish
+ *   - numeric remaining quota: NO CLI/api-key reachable API
+ *   - per-request usage.credit in chat SSE body
+ *
+ * What the inventory lacks (this probe's target): a FULL scan across every
+ * reachable endpoint of BOTH response headers AND body fields for quota
+ * signals, incl. agenttool bodies, error envelopes, and dosage-notify
+ * parameter sensitivity.
+ *
+ * Pre-registered predictions (2026-08-19):
+ *   P-Q1  accounts, /v3/config, dosage-notify, and a 12403 error envelope
+ *         all carry ZERO quota-ish RESPONSE HEADERS (the chat scan
+ *         generalizes: this gateway puts no quota in headers at all)
+ *   P-Q2a agenttool search body has NO usage/credit field (the UA-probe
+ *         snippet showed {query,type,provider,results} only)
+ *   P-Q2b agenttool webfetch body HAS a usage field (index.js:653 records
+ *         data?.usage conditionally — predicts it exists)
+ *   P-Q3  dosage-notify ignores parameters: {threshold:100} returns the
+ *         same healthy shape as {} (dosageNotifyCode 0, empty strings)
+ *   P-Q4  the 12403 error response (v3/config with a gate-failing UA)
+ *         carries zero quota-ish headers — errors don't leak quota either
+ * Exploratory: GET /v2/report existence (CLI endpoint list mentions it).
+ *
+ * Low rate: 2s spacing, single account, read-only (search max_results=1,
+ * webfetch example.com, chat max_tokens=1).
+ *
+ * Usage: node scripts/probe-quota2.mjs [--out docs/probes/quota-map-YYYY-MM-DD.jsonl]
+ */
+
+import { readFileSync, existsSync, appendFileSync, mkdirSync } from 'node:fs'
+import { homedir } from 'node:os'
+import { join, dirname } from 'node:path'
+
+const DSH_HOME = process.env.DSH_HOME ?? join(homedir(), '.dsh')
+const GATEWAY = 'https://copilot.tencent.com'
+const UA = 'CLI/unknown CodeBuddy/2.136.0'
+const SPACING_MS = 2000
+const OUT = process.argv.includes('--out')
+  ? process.argv[process.argv.indexOf('--out') + 1]
+  : join('docs', 'probes', `quota-map-${new Date().toISOString().slice(0, 10)}.jsonl`)
+
+const QUOTA_RE = /quota|credit|limit|remain|balance|dosage|usage|rate|plan/i
+
+function resolveCredential() {
+  try {
+    const cfg = JSON.parse(readFileSync(join(DSH_HOME, 'codebuddy-plugin.json'), 'utf8'))
+    if (cfg.authMode === 'oauth') {
+      const auth = JSON.parse(readFileSync(join(DSH_HOME, 'codebuddy-plugin-auth.json'), 'utf8'))
+      if (auth?.auth?.accessToken) return { authorization: `Bearer ${auth.auth.accessToken}`, apiKey: null }
+    }
+    const active = (cfg.apiKeys ?? []).find((k) => k.name === cfg.activeApiKey)
+    if (active?.key) return { authorization: `Bearer ${active.key}`, apiKey: active.key }
+  } catch { /* fall through */ }
+  if (process.env.CODEBUDDY_API_KEY) return { authorization: `Bearer ${process.env.CODEBUDDY_API_KEY}`, apiKey: process.env.CODEBUDDY_API_KEY }
+  const credFile = join(DSH_HOME, '.credentials.yaml')
+  if (existsSync(credFile)) {
+    const m = readFileSync(credFile, 'utf8').match(/^\s*CODEBUDDY_API_KEY:\s*["']?([^"'\s]+)["']?\s*$/m)
+    if (m) return { authorization: `Bearer ${m[1]}`, apiKey: m[1] }
+  }
+  throw new Error('no CodeBuddy credential found')
+}
+
+const sleep = (ms) => new Promise((r) => setTimeout(r, ms))
+
+/** Deep-scan a parsed body for quota-ish KEY NAMES; returns paths. */
+function bodyHits(value, path = '', hits = []) {
+  if (Array.isArray(value)) value.forEach((v, i) => bodyHits(v, `${path}[${i}]`, hits))
+  else if (value && typeof value === 'object') {
+    for (const [k, v] of Object.entries(value)) {
+      const p = path ? `${path}.${k}` : k
+      if (QUOTA_RE.test(k)) hits.push(p)
+      bodyHits(v, p, hits)
+    }
+  }
+  return hits
+}
+
+const redact = (v) => typeof v === 'string'
+  ? v.replace(/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}/gi, '<redacted>').replace(/\b\d{6,}\b/g, '<redacted>')
+  : v
+
+async function probe(cred, arm) {
+  const headers = { Accept: 'application/json', Authorization: cred.authorization, 'User-Agent': arm.ua ?? UA }
+  if (cred.apiKey) headers['x-api-key'] = cred.apiKey
+  if (arm.body) headers['Content-Type'] = 'application/json'
+  const t0 = Date.now()
+  try {
+    const res = await fetch(`${GATEWAY}${arm.path}`, {
+      method: arm.method ?? (arm.body ? 'POST' : 'GET'),
+      headers,
+      body: arm.body ? JSON.stringify(arm.body) : undefined,
+    })
+    const resHeaders = {}
+    res.headers.forEach((v, k) => { resHeaders[k] = v.length > 80 ? v.slice(0, 80) + '…' : v })
+    const text = await res.text()
+    let json = null
+    try { json = JSON.parse(text) } catch { /* SSE or plain text */ }
+    return {
+      name: arm.name, path: arm.path, pred: arm.pred, note: arm.note,
+      status: res.status,
+      headerNames: Object.keys(resHeaders),
+      quotaHeaders: Object.keys(resHeaders).filter((k) => QUOTA_RE.test(k)),
+      code: json?.code ?? null,
+      bodyQuotaPaths: json ? bodyHits(json) : [],
+      bodySample: redact(json ? JSON.stringify(json).slice(0, 300) : text.slice(0, 200)),
+      ms: Date.now() - t0,
+    }
+  } catch (err) {
+    return { name: arm.name, path: arm.path, pred: arm.pred, note: arm.note, error: String(err?.message ?? err), ms: Date.now() - t0 }
+  }
+}
+
+const ARMS = [
+  { name: 'P-Q1a-accounts', path: '/v2/accounts', pred: 'P-Q1', note: 'zero quota-ish response headers' },
+  { name: 'P-Q1b-config', path: '/v3/config', pred: 'P-Q1', note: 'zero quota-ish response headers' },
+  { name: 'P-Q1c-dosage', path: '/v2/billing/meter/get-dosage-notify', body: {}, pred: 'P-Q1', note: 'zero quota-ish response headers' },
+  { name: 'P-Q4-12403-headers', path: '/v3/config', ua: 'Garbage/0.0', pred: 'P-Q4', note: 'error envelope: zero quota headers' },
+  { name: 'P-Q2a-search-body', path: '/agenttool/v1/search', body: { query: 'kimi', type: 'text2text', max_results: 1 }, pred: 'P-Q2a', note: 'search body has NO usage/credit field' },
+  { name: 'P-Q2b-webfetch-body', path: '/agenttool/v1/webfetch', body: { url: 'https://example.com/' }, pred: 'P-Q2b', note: 'webfetch body HAS usage field' },
+  { name: 'P-Q3-dosage-params', path: '/v2/billing/meter/get-dosage-notify', body: { threshold: 100, level: 'high' }, pred: 'P-Q3', note: 'params ignored: same healthy shape as {}' },
+  { name: 'X-report-exists', path: '/v2/report', pred: null, note: 'exploratory: CLI endpoint existence' },
+]
+
+const cred = resolveCredential()
+mkdirSync(dirname(OUT), { recursive: true })
+appendFileSync(OUT, JSON.stringify({ type: 'run', at: new Date().toISOString(), gateway: GATEWAY }) + '\n')
+console.log(`evidence → ${OUT}`)
+
+for (const arm of ARMS) {
+  const rec = { type: 'probe', ...await probe(cred, arm) }
+  appendFileSync(OUT, JSON.stringify(rec) + '\n')
+  console.log(`${arm.name.padEnd(22)} status=${rec.status ?? '-'} code=${rec.code ?? '-'} quotaHeaders=${(rec.quotaHeaders ?? []).length} bodyQuotaPaths=${JSON.stringify(rec.bodyQuotaPaths ?? [])} ${(rec.bodySample ?? rec.error ?? '').slice(0, 100)}`)
+  await sleep(SPACING_MS)
+}
+console.log('done.')