thincoder 0.12.59 → 0.12.60

package/CHANGELOG.md

@@ -1,6 +1,41 @@
+## [0.12.60] — 2026-09-08
+
+> 0.12.59 → 0.12.60（§4.5 连续号——发布时定号）
+
+### Changed
+
+- **R17 会诊/飞刀完全异步化（AGENT-LOOP.md §25——2026-09-06/07）**：consult_check 退役（digest 自动注入闭环）；escalate 入 other 池异步化（async:false 保留）；三族 pending 注入/驱动推广；T-R17a..r 全绿（CLI L2 1626/1626 + VS 1301/1301）
+- **记账路径 join→resolve 双前缀修复（R-bug——2026-09-07——AGENT-LOOP.md §24 尾修复注）**：record-results.mjs ×2 + agent.mjs ×1 `join(agent.cwd, p)` → `resolve(agent.cwd, p)`（绝对路径 p 双前缀——touchedFiles/记账错路径——修复前先验失败实拍——guards.test +12 断言）
+- **R18 长测试输出先落盘纪律条款（AGENT-LOOP.md §26——2026-09-06 快车道）**：`src/prompts/system.md` Rules 工具纪律区插逐字条款（"长输出命令先落盘"——全量/长测试（≥60s）先重定向日志再查——汇总读尾部/失败 grep——禁过滤管道直跑长命令——镜像锚逐字稿照抄 AGENT-LOOP §26.2——各端独立断言）+ 日志位置执行细则（OS 临时/~/.thincoder 类非工作区——查毕删除——防 git 工作区 untracked 污染——条款逐字不改）。锚测试 T-R18.1/T-R18.2（fail-when-unchanged）。双端各自落端（VS Code 侧并行照抄）。
+- **R19 跨会话历史检索 + 检索族消歧（SESSION.md §13——2026-09-06）**：`read_history` 加 `path` 参数（可选——缺省 = 本会话零行为变化）——path = 会话文件路径 → 跨会话单槽深查（同 filter 面 role/keyword/tool/since/until/limit/direction 应用 + limit clamp）；path = "cwd:xxx" → 发现面（列该 cwd 全部槽——每槽摘要行含槽号 + 完整文件路径 + title/消息数/updatedAt——不做死槽过滤——时间序）；行扫护栏 READ_HISTORY_SCAN_MAX = 200,000 行——超限返回定稿文案 `{"error":"session too large — refine keyword or since/until"}`（不再读全文）；read_history 描述尾段插族表总纲（D-R19b 逐字——checkpoint cat/versions 经核验保留——git 动作面含 versions）；recent_changes/memory/doc_search/code_search 描述补互指句。测试 T-R19.1..7（含 3b/3c/3d 错误/边界子行 + 描述锚）。压缩刀锋 fixture 随 schema 增重校准（session-compaction T3b 15500 → 17000——沿革注更新）。
+
+- **legacy v1 扩展删除集（SESSION.md §12.7——2026-09-06）**：`session gc` 冷 cwd 整前缀删除集补裸 `{hash}.json`（v1 死数据——§12.6 裁决代码同步）；三处同源单集合（dry-run 报告/confirm 警告/删除共用）；自动残留 GC 面零触碰。测试 T-V1a..d（含 v1 报告/删除/--all/活跃拒绝）。父侧 L2 1543/1543 全绿。
+
+- **测试分层 Phase 3：存量清理 + 红线收窄（TESTING.md §1 D-T4/D-T5——2026-09-06——用户逐项批准删除清单）**：删除一次性迁移零残留锁 11 组 + 防回潮元测试 4 案 + 同锚多断 3 组（1492→1486，差额 -6 全对账）；"断言数不减"红线收窄为拆分轮专用（system.md Module Split Policy + 双 METHODOLOGY.md 拆后两验注——清理轮改删除清单制）。测试 T-T5 收窄锚断言。**五杠杆批全链收口——总目标 AC-T4 ✓：链测试墙钟 ~4.5-5.5min → ~1.6min（降 ~70%）**。
+
+- **测试分层 Phase 2：eng-coder 首次实现降 L0+（TESTING.md §1 D-T3——2026-09-06——supersede §18.7 D-TS1 首次实现全量 L1 口径）**：`engineering-sub.md` 验证粒度段修订——首次实现 = L0+（语法检查 + §18.12 定向相关测试，秒级），L1 收窄为显式升级档（兜底语义保留），修正轮 L0 与父侧 L2 不变；交付报告义务锚句（"not full-suite verified — the parent-side L2 run is the only full-suite point."——双端逐字镜像）。AGENT-LOOP §18.7 原位 supersede 注。测试 T-T4 锚断言（fail-when-unchanged + 旧句零残留）。实测快层 18.5s 全绿 1492 用例。
+
+- **测试分层 Phase 1：slow 门防漏拦截 + 快层归册（TESTING.md §1——2026-09-06——用户"测试已成包袱"五杠杆批之零冲突先行段）**：`test/slow-gate.mjs`（reporter 收集叶子用例耗时——未标 slow 超 800ms 拦截阈值即报错红，防快层腐化——env `THINCODER_SLOW_GATE_MS` 可覆盖）+ `test/run-fast.mjs`（npm test 新入口——spec 直通 + gate 判定非零退出 + `--test-concurrency=6`）；60 条重 IO 用例归册（只加 slow() 标记，零删除零本体改动）；机制自验 `test/slow-gate.test.mjs` 红/绿双证。**实测：快层 38.0s → 19.2-19.7s（≤20s 达标），test:full 1491/1491 全绿**。
+
+- **拆分判据修辞防博弈加固（METHODOLOGY.md 2026-09-06 变更条——快车道——用户实测模型引用"才 500 多行、没 2300 行大"拒绝拆分）**：文件档升同级判定表（>500 = 必须拆——硬限、无例外、无相对比较通道——原"兜底红线"定性退役）；函数档改定位 = 早期预警线、文件档 = 不可协商底线；新增"相对大小永远不是豁免"封堵句（中英镜像）+ 动机节封口句 + 反例档案新行（相对大小打折硬限实测反例）；**AGENTS.md / .thincoder/advisor.md 测试文件+生成代码豁免说辞全删（用户裁定"不允许豁免"——工作区根两文件同删）**——存量 26 个 >500 测试文件登记 TODO 拆分轮。测试 T-S1/S2/S3/S5 锚断言（prompts.test.mjs——含旧修辞零残留负向断言）。
+
+- **question 工具使用抑制（AGENT-LOOP.md §22——2026-09-06 快车道——用户实测过度提问/大段文字/一条多问）**：工具描述 Notes 补三条逐字锚（一次一问不塞子问题 / 问题文本要短、背景走正文回复 / 例行确认门走普通文本回复、question 仅留真实决策场景）；提示词三文件锚（engineering.md Questioning Style 两条 + system.md 确认门形式句 + discipline.md 工具表反模式扩展）；questionTool execute 机械上限（git.mjs 寄生段——question >100 字符 / options >4 条返回错误串、不弹卡不调 onQuestion——用户裁定值——guard 仅 `Array.isArray(options) && length > 4` 触发、非数组透传）。测试 T-Q1/T-Q3/T-Q4/T-Q5/T-Q6/T-Q7（git.test + prompts.test——锚 fail-when-unchanged + 超限拒绝/边界透传）。
+
+- **async spawn 收尾引导（AGENT-LOOP.md §19.7——2026-09-06）**：subagent 工具描述 Async spawn 段尾部补收尾引导锚句（"After an async spawn the turn winds down normally — nothing expects you to wait for it…"——逐字定稿于设计文档 §19.7 D-A2）——模型 async spawn 后回合自然收尾、unchecked 结果自动到达（下轮注入/挂起态 digest）、无需自发 status/check 轮询等待；锚句内容断言 fail-when-unchanged（设计锚 + 两端各自断言——防描述再漂移——D-A4）。测试 T-D-A2。
+- **subagent check 动作删除 + 锚句收紧（AGENT-LOOP.md §19.8——2026-09-06 用户裁定"check 是冗余 API"）**：工具面六动作 → 五动作（spawn/status/escalate/cancel/panel）——删 `action:"check"` + `n` 参数 + `executeCheckAction`/`MAX_ASYNC_CHECKS`/`_asyncCheckLastN` 全链路；async 子代理结果仅自动通道送达（回合尾 collect/挂起 digest——§17.5 机制零改动——consumed 终态墓碑保留——dependsOn 语义不变）；Async spawn 段描述替换为 D-CH2 锚句逐字（"…do not poll or wait for the result…use a synchronous spawn instead—pass `async:false`…"——fail-when-unchanged 断言）；main.md/discipline.md/工具描述 check 引导句改自动送达口径；停滞检测（P-SL2）check 守卫路径删除——F-SL2 语义面迁至 status 视图（可显示停滞标记不变）。测试：check 域用例删 8 条（逐条见交付报告）+ 2 条在位改自动通道形态（T-SD14/T-H3）+ T-M16 改 status 形态 + D-A2 断言在位改 D-CH2。
+- **工程模式 token 防伪层删除（ENGINEERING-MODE.md 2026-09-06 段——用户裁定"过度工程给我删了"）**：advisor.mjs 删 DEFAULT_TOKEN_SECRET / USING_DEFAULT_SECRET / TOKEN_SECRET / warnIfDefaultSecret + 启动警告与 HMAC 签名/验签——token 回归**无签名流程凭证**（格式 uuid:expiresAt——TTL 检查 fail-closed + designId 槽位精确匹配不变）；存量 3 段 token 格式判错 fail-closed 拒绝（不迁移——重评一次）；NFR3/§2.3/§2.7/§2.6 文档同步。
+- **会话"当前槽位"按端分离（R4——2026-09-05——SESSION.md §10）**：CLI 与 VS Code 各自 end marker（`{manifest}.cli|.vscode`，本端单写者）记录"本端最后使用的槽位"——恢复决策 resumeSlot（本端记录 → 无记录一次性继承 active 死主槽 → 全新分配）替代共享 manifest.active 作为恢复依据——双端同开时退出 CLI 重进不再进错会话/空白新槽；启动即钉 _slot 闭合"load→首保存"迁移窗口；/session 列表高亮按本端记录（含被删守卫）。双端同批镜像（L2：CLI 1485/1437+48skip/0 + VS Code 1196/1196 全绿）。
+- **git 工具 add/rm/commit 多路径（2026-09-05 发版后——git add 单路径被迫 N 次调用）**：`path` 空格分隔多路径（`git add a b c`——ref 多值先例同法——split(/\s+/)）；path 描述尾追加注（保既有原子串断言）；双端同改（CLI git.test + VS Code git.test 镜像用例——add/rm/commit 三动作多路径覆盖）。
+- **release:check 一键发版门禁（scripts/release-check.mjs——2026-09-05 0.12.59 发版教训）**：lint + test:full 合并一键；全量输出捕获不刷屏（只打摘要）；失败自动提取 failing tests 详情段（此前为看错误详情跑 3 次全量）；RELEASE.md §1 发布前检查合并 + 修复迭代局部重跑纪律 + prepublishOnly 双保险注。
+
+### Fixed
+
+- fix: 子代理块 done 后定格修复批（2026-09-07 id:14）——CLI trimSubTree done 豁免 + route* done 守卫（迟到 chunk 丢弃）
+- fix: prompts.test 折行断言折叠化收口（2026-09-07 id:3）——双侧 \s+ 归一匹配——负守卫保持负向——锚整块/落位序断言加固
+
 ## [0.12.59] — 2026-09-05
 
-> 0.12.58 → 0.12.59（§1.5 连续号——发布时定号）
+> 0.12.58 → 0.12.59（§4.5 连续号——发布时定号）
 
 ### Added
 
@@ -16,7 +51,7 @@
 - **普通模式两段式（AGENT-LOOP.md §21 F-N1.5——2026-09-05 用户"把 coder 用起来……解决自查问题"）**：规模实现批次默认委托 coder 子代理（async——设计书为 task book）——执行/检查心智分离（隔离上下文破自查盲区）；小改动/探索留内联；复核走 F-N1.4（偏差退回 coder ≤2 轮）。测试 T-N1.9 双端。
 - **委托操作标准（AGENT-LOOP.md §21 扩展注 4——F-N1.6——2026-09-05 用户"spawn coder 干活现在并没有明确的标准是吗"）**：规模判据可操作化（≥2 文件/单文件 >30 行逻辑/模块边界/双端镜像 → 委托；≤30 行/文档同步/探索 → 内联）；任务书七字段标准（目标/已知事实/设计+禁止/约束/硬验收/报告格式/调度元数据——缺字段=委托缺陷）；委托模式判据（async 默认/sync 仅依赖链/并行仅文件域互斥）；通用验收基线。测试 T-N1.10 双端。
 - **/config 交互修复（2026-09-05）**：改配置项保存后回主菜单（不再退到输入框）——每轮刷新 ac/tc 配置引用（reloadConfig 换对象后显示新值）；view 浏览态同回菜单。
-- **版本号连续性规则（RELEASE.md §1.5——2026-09-05 用户裁定"不要跳号"）**：号在发布时定、开发期不预占（CHANGELOG 挂 [Unreleased]）；待发号 = registry 最高 + 1；缺口不补。
+- **版本号连续性规则（RELEASE.md §4.2/§4.3——2026-09-05 用户裁定"不要跳号"）**：号在发布时定、开发期不预占（CHANGELOG 挂 [Unreleased]）；待发号 = registry 最高 + 1；缺口不补。
 
 ### Fixed
 
@@ -80,7 +115,7 @@
 
 ### Added
 
-- **`/mcp` edit/add 统一字段 picker 表单（v2）+ agent 代配闭环**（docs/design/MCP.md §5，CLI-only）：① **edit = 字段选择表单**——picker 列可编辑字段行（`HTTP URL https://…` / `Token d90c26bb…` 打码 / `Headers 2 items`；stdio `Command/Args/Env`；name 不可改无行）+ 末行 `✓ Save & test`；选中字段只输入该字段新值（提示 `(current: …)`——空=不变、`-`=删可选字段、`k=`=删 header/env 项、required 字段拒绝 `-`）→ 回 picker 循环连改多字段（中间 Esc 回 picker 不丢已改值）；**废除逐字段预填重问**（改 token 不再被迫路过 URL/headers）；② **add 复用同一表单**——空 entry 起、必填字段 `(required)` 标注、Save 校验必填非空（未满足提示并停留表单不落盘）、headers/env 不选即跳过、add 的 name 可编辑（重复名检查）；③ **保存前预览 + 探活 + 字段级重试合一**——`showPreview`（token 遮蔽）→ `probeMcpServer` 零副作用探活（`✓ N tools, Xms` / `✗ 错误`）→ `Save? (Y/n)`；探活失败 → `Save anyway? (y/N)` 显式 y，否则**回同一字段表单**只重输失败字段复 probe（**独立 retry 路径废除**，不重启流程；save-anyway 显式 y 保留）；④ **表单文件拆分**——新增 `src/tui/cmd-mcp-form.mjs`（fieldPicker 机制独立文件；`cmd-mcp.mjs` 499→382 行，脱离 500 硬限压线）；⑤ **列表即菜单**——主菜单 = server 行（●/○ 连接态 + tool 数）+ `+ Add server` + `↻ Refresh` + 顶部 agent 代配提示行；选中行 → per-server 子菜单（Edit/Test/Reconnect/Remove），"先选操作再选 server"双弹层与 View list 废除；⑥ **磁盘重读**——`config.mjs` 新增 `reloadMcpFromDisk()`（菜单打开边界 + Refresh）：磁盘→内存仅替换 mcp 段；畸形 config.json 回退内存态 + `⚠ disk config unreadable` 提示行；disk 删除/变更的已连接 server 连接不断 + 行尾 `⚠ disk changed` 对账标记（persistRaw 落盘后重读幂等防环）；⑦ AI 生成降 transport picker 末位（生成的 entry 同走预览+探活确认环，失败回表单补齐/修正）；`/mcp edit|test|remove|connect <name>` 直达参数路径零改动
+- **`/mcp` edit/add 统一字段 picker 表单（v2）+ agent 代配闭环**（docs/design/MCP.md §5，CLI-only）：① **edit = 字段选择表单**——picker 列可编辑字段行（`HTTP URL https://…` / `Token ••••…` 打码 / `Headers 2 items`；stdio `Command/Args/Env`；name 不可改无行）+ 末行 `✓ Save & test`；选中字段只输入该字段新值（提示 `(current: …)`——空=不变、`-`=删可选字段、`k=`=删 header/env 项、required 字段拒绝 `-`）→ 回 picker 循环连改多字段（中间 Esc 回 picker 不丢已改值）；**废除逐字段预填重问**（改 token 不再被迫路过 URL/headers）；② **add 复用同一表单**——空 entry 起、必填字段 `(required)` 标注、Save 校验必填非空（未满足提示并停留表单不落盘）、headers/env 不选即跳过、add 的 name 可编辑（重复名检查）；③ **保存前预览 + 探活 + 字段级重试合一**——`showPreview`（token 遮蔽）→ `probeMcpServer` 零副作用探活（`✓ N tools, Xms` / `✗ 错误`）→ `Save? (Y/n)`；探活失败 → `Save anyway? (y/N)` 显式 y，否则**回同一字段表单**只重输失败字段复 probe（**独立 retry 路径废除**，不重启流程；save-anyway 显式 y 保留）；④ **表单文件拆分**——新增 `src/tui/cmd-mcp-form.mjs`（fieldPicker 机制独立文件；`cmd-mcp.mjs` 499→382 行，脱离 500 硬限压线）；⑤ **列表即菜单**——主菜单 = server 行（●/○ 连接态 + tool 数）+ `+ Add server` + `↻ Refresh` + 顶部 agent 代配提示行；选中行 → per-server 子菜单（Edit/Test/Reconnect/Remove），"先选操作再选 server"双弹层与 View list 废除；⑥ **磁盘重读**——`config.mjs` 新增 `reloadMcpFromDisk()`（菜单打开边界 + Refresh）：磁盘→内存仅替换 mcp 段；畸形 config.json 回退内存态 + `⚠ disk config unreadable` 提示行；disk 删除/变更的已连接 server 连接不断 + 行尾 `⚠ disk changed` 对账标记（persistRaw 落盘后重读幂等防环）；⑦ AI 生成降 transport picker 末位（生成的 entry 同走预览+探活确认环，失败回表单补齐/修正）；`/mcp edit|test|remove|connect <name>` 直达参数路径零改动
 - **MCP Streamable POST 误判修复 + `/mcp edit`/`/mcp test` + token 一等字段**（docs/design/MCP.md §4，两端落地）：① `httpTransport` 增 postOnly 标记——GET SSE 405 降级后的纯 POST 模式 `isAlive()` 不再因 `eventSource == null` 误判死（glm-websearch "reconnect failed after 4 attempts" 根因：降级后 isAlive 恒 false → ensureAlive 触发无意义重连循环；legacy SSE 流断仍正常 fireDead 重连不回归）；② `/mcp edit [name]`：逐字段预填重问（空输入保留 / `-` 删除可选字段 / `k=` 删除单个 header 项），persistRaw 原位替换保数组序，保存后自动重连（config 指纹含 token，变更自动关旧连接）；③ `/mcp test [name]`：probeMcpServer 一次性探活（initialize + tools/list 计时 → `OK — N tools, Xms` / 错误透传），零副作用（不进 session 表、不动 agent.tools、探完即关）；④ token 一等字段：config 增 `token` 字段，connect 链自动合成 `Authorization: Bearer <token>`（显式 headers 优先，不写回 config）；⑤ parseHeaders 改逗号分隔（`Authorization=Bearer abc, X-Foo=bar`——修复空格截断把 Bearer token 截成 "Bearer" 的缺陷）；⑥ VS Code 同构：http transport 同款修复 + probeMcpServer 镜像 + 面板 [Edit]/[Test] 按钮（同一表单编辑预填，token 字段 + 逗号分隔 headers 提示）+ **面板 [Reconnect] 死按钮修复**（webview 发的 `reconnectMcp` 消息在路由拆分时丢失 case，按钮此前无效）
 - **memory_delete 工具**（跨端）：三层记忆条目删除——personal 行级删（embedding/FTS 随行）、project/team 文件级删；scope 与 id 前缀匹配校验，非法 scope 明确报错
 

package/README.md

@@ -15,13 +15,13 @@ Design philosophy (the entire meaning of the name): if the Node standard library
 - **Fix-verify loop**: file changes without `verify` get pushed back — syntax check + tests must pass before the agent can claim completion (auto-repair up to 3 rounds)
 - **Checkpoint system**: auto-snapshot before every user task, `list`/`create`/`rewind` tools for the model, single-file restore — rewinding itself is reversible (pre-rewind state auto-saved)
 - **Codebase understanding** ⭐0.5.0: `repo_outline` (dependency outline, auto-injected at startup), `code_search` (source FTS5 + vectors + JSDoc extraction), `doc_search` (docs chunked by ## headings) — background indexing, auto-incremental updates on file writes, three tools guided by "structure → intent → details"
-- **Model adaptation** ⭐: top-tier only, latest only. Built-in flagship models from seventeen providers — DeepSeek / Kimi / Kimi For Coding / GLM / Qwen / Qwen Token Plan / MiniMax / OpenAI / Claude / Gemini / Grok / Mistral / Volcengine Ark (豆包) / Hunyuan (腾讯混元) / SiliconFlow (硅基流动) / OpenRouter / Groq. No legacy model compatibility. Auto-matched context windows, truncation-resume protocols (prefix/partial), thinking-mode APIs (thinking.type / reasoning_effort), reasoning_content echo strategies (reasoningEcho), output limits, temperature range clamping — all deeply adapted.
+- **Model adaptation** ⭐: top-tier only, latest only. Built-in flagship models from twenty providers — DeepSeek / Kimi / Kimi For Coding / GLM / GLM Coding Plan (智谱 Coding) / Qwen / Qwen Token Plan / MiMo (小米) / MiMo Token Plan (小米套餐) / MiniMax / OpenAI / Claude / Gemini / Grok / Mistral / Volcengine Ark (豆包) / Hunyuan (腾讯混元) / SiliconFlow (硅基流动) / OpenRouter / Groq. No legacy model compatibility. Auto-matched context windows, truncation-resume protocols (prefix/partial), thinking-mode APIs (thinking.type / reasoning_effort), reasoning_content echo strategies (reasoningEcho), output limits, temperature range clamping — all deeply adapted.
 - **Toolset**: `read` / `write` / `edit` / `bash` / `glob` (supports `**`) / `grep` / `websearch` / `ls` / `fetch` + `read_image` (image/video paste) + three retrieval tools + MCP — all zero-dependency, file tools confined to the working directory
 - **Memory system**: three layers (personal/project/team), FTS5 + vector RRF hybrid retrieval, git-friendly markdown format
 - **Two-phase tool scheduling**: permission prompts serialized, read-only tools parallelized, side-effect tools serialized
 - **Session persistence** ⭐0.5.0: unlimited archive slots, `/session` to switch anytime, tool results visible after restore. Process-level isolation — multiple instances in the same directory each get their own session slot
 - **Concurrent subagents**: three roles — `explore`/`plan`/`coder` — dispatched in parallel, streaming output visible, reports land in the conversation; per-subagent model override (`subagent` tool `model` arg or `agent.subagentModel` config — e.g. discuss with `glm-5.2`, let `deepseek-v4-flash` implement)
-- **Multi-model consultation 会诊 + 飞刀 (escalate)** ⭐0.12.30: `consult_start`/`consult_check`/`consult_stop` run several configured models in parallel as independent read-only consultants (each with its own TUI activity card, `main_history` access to the failure trail); `escalate` flies in a stronger model for a single expert implementation run with full write access. Candidate pool = `agent.consultModels` ([{ provider, model, effort? }], up to 5); budgets via `agent.consultTurns` / `agent.consultTimeoutMs`
+- **Multi-model consultation 会诊 + 飞刀 (escalate)** ⭐0.12.30: `consult_start`/`consult_stop` run several configured models in parallel as independent read-only consultants (each with its own TUI activity card, `main_history` access to the failure trail); when every model settles, the full verdict digest is delivered automatically (R17 — consult_check retired). `escalate` flies in a stronger model for a single expert implementation run with full write access (default-async since R17 — the post-op report arrives automatically with its mutations merged). Candidate pool = `agent.consultModels` ([{ provider, model, effort? }], up to 5); budgets via `agent.consultTurns` / `agent.consultTimeoutMs`
 - **Plan Mode**: read-only exploration + design, implement after user approval
 - **AUTO mode**: `/auto` full authorization, no confirmations on long tasks
 - **Task tracking**: `task` tool breaks down multi-step work, status bar ✓n/m live progress, auto-filters completed items

package/bin/thincoder.mjs

@@ -7,6 +7,7 @@
  *   thincoder memory <sub>    Memory management: list / search / put / remove
  *   thincoder upgrade         Update to the latest version from npm
  *   thincoder completion <sh> Shell completion: bash / zsh / fish
+ *   thincoder session gc    Session dir GC: --dry-run report / --confirm delete cold projects (SESSION.md §12)
  *   thincoder acp             Agent Client Protocol server (stdio, for Zed/JetBrains/Paseo)
  *   thincoder -v              Print version
  *   thincoder --help          Print help
@@ -23,19 +24,56 @@ import { memoryCommand } from "../src/cli/memory-command.mjs"
 import { setupWizard } from "../src/cli/setup-wizard.mjs"
 import { summarize, askPermission } from "../src/cli/permission.mjs"
 import { distillCommand } from "../src/cli/distill-command.mjs"
+import { prepareCrashReporting, recentCrashHint, writeCrashRecord } from "../src/crash-reports.mjs"
+import { setTuiActive, restoreTerminalAfterCrash } from "../src/tui/tui-lifecycle.mjs"
 
 const [command, ...args] = process.argv.slice(2)
 const VERSION = JSON.parse(readFileSync(new URL("../package.json", import.meta.url), "utf8")).version
 
-// Top-level safety net: print one-line error and exit cleanly, no stack traces to the user
-process.on("uncaughtException", (error) => {
-  console.error(`[error] ${error.message}`)
-  exitSoon(1)
-})
-process.on("unhandledRejection", (error) => {
-  console.error(`[error] ${error?.message ?? error}`)
+// R25（F-R25b）：crash-reports 预建 + process.report 启用——入口最前（一切重活前——缩编程
+// 期窗口）——V8 OOM/原生 fatal 自动写 report.*.json（实现批实测：目录缺失时 Node 静默不写
+// ——预建为必要动作）。失败不阻断启动（尽力面）。
+prepareCrashReporting()
+
+// R25（F-R25a）异常钩子升级：原"只 console.error 一行（TUI 全屏下不可见）"→ ① 落盘
+// crash-{ts}-{pid}.json ② TUI 活动态终端恢复 ③ console.error（恢复后打印才可见）
+// ④ exit 非 0。执行序列定死（评审 #5 + 复审 #2）：一次性 guard（防再入递归）→ 同步写 →
+// 终端恢复 → console.error → exit 非 0——各步独立 try/catch（写失败/恢复失败不阻断后续步）。
+let crashHandled = false
+function handleFatal(type, error) {
+  if (crashHandled) {
+    // 再入（本序列内又抛异常）→ 直接退出，不再递归
+    process.exit(1)
+  }
+  crashHandled = true
+  try {
+    // ① 同步写诊断记录（设计字段集：时间/类型/消息+堆栈/uptime/argv/cwd/内存/node+版本）
+    writeCrashRecord({ type, error })
+  } catch { /* 写失败不阻断后续步 */ }
+  try {
+    // ② TUI 终端恢复：仅 TUI 活动态执行（writeCleanupSequence 同源——chat 模式 stdout
+    // 管道不得收 ANSI——误判即污染输出）
+    restoreTerminalAfterCrash()
+  } catch { /* 恢复失败不阻断后续步 */ }
+  try {
+    // ③ 原 console.error 行保留（非 TUI 模式可见——恢复后打印才可见）
+    console.error(`[error] ${error?.message ?? error}`)
+  } catch { /* 打印失败不阻断 exit */ }
+  // ④ exit 非 0 恒达（exitSoon——Windows/Node 24 上 fetch 后立即 exit 触发 libuv 断言先例）
   exitSoon(1)
-})
+}
+
+// Top-level safety net: print one-line error and exit cleanly, no stack traces to the user
+process.on("uncaughtException", (error) => handleFatal("uncaughtException", error))
+process.on("unhandledRejection", (error) => handleFatal("unhandledRejection", error))
+
+// R25 测试门（T-R25a.1/a.2——子进程 env 注入——生产零路径）：THINCODER_TEST_CRASH=1 抛
+// 未捕获异常走完整崩溃序列；THINCODER_TEST_TUI_ACTIVE=1 模拟 TUI 活动态（同一 setter——
+// 测试缝同源）；THINCODER_TEST_CLEANUP_OUT 指向文件时恢复序列写入该文件（tui-lifecycle 读）。
+if (process.env.THINCODER_TEST_CRASH === "1") {
+  if (process.env.THINCODER_TEST_TUI_ACTIVE === "1") setTuiActive(true)
+  throw new Error("R25 test crash (THINCODER_TEST_CRASH)")
+}
 
 const USAGE = `thincoder - thin coding agent
 
@@ -49,9 +87,11 @@ Usage:
   thincoder memory remove <id>                 Remove an entry
   thincoder sync              Sync team memory repo (pull --rebase + reindex)
   thincoder reindex           Rebuild the local index from markdown sources
-  thincoder distill <file> [--yes] [--scope=<s>]
+  thincoder distill <file> [--yes] [--layer=<s>]
                             Extract knowledge candidates from a session
                             transcript file; confirm each before saving
+  thincoder session gc --dry-run | --confirm <hash|--all>
+                            Session dir GC: report/delete cold project data (cold = manifest idle >90d, no live slots)
   thincoder upgrade         Update to the latest version from npm
   thincoder completion <sh>  Generate shell completion script (bash / zsh / fish)
   thincoder -v, --version   Print version
@@ -86,6 +126,10 @@ switch (command) {
       break
     }
 
+    // R25（F-R25c）：非交互/chat 模式——上次异常终止提示写 stderr 一行（无匹配不提示）
+    const crashNotice = recentCrashHint()
+    if (crashNotice) console.error(crashNotice)
+
     const agent = await assembleAgent()
     // SESSION.md §8 D-S4（F4）：headless 无 TUI —— 可读错误 + 退出码 1，不弹 UI、不崩溃
     if (agent._providerInvalid) {
@@ -239,10 +283,12 @@ switch (command) {
     if (agent._providerInvalid) agent.provider = null
     const config = loadConfig()
     // 恢复上次的会话（同一项目目录）；provider 按保存的名字切回（用户上次可能换过模型）
-    const { loadSession, applySession } = await import("../src/session.mjs")
-    const restored = loadSession(process.cwd())
-    if (restored) {
-      const switched = applySession(agent, restored)
+    // 2026-09-05 §10（R4）：恢复决策按本端记录 resumeSlot（D-2 ①②③）——manifest active
+    // 只作"无记录端"的一次性继承源，不再作本端恢复第一依据（D-6）。
+    const { resumeSlot, applySession } = await import("../src/session.mjs")
+    const { slot, data } = resumeSlot(process.cwd())
+    if (data) {
+      const switched = applySession(agent, data)
       if (switched && agent.config?.agent?.compactThresholdAuto) {
         // 压缩阈值跟模型走（与 TUI 切换 provider 时的处理一致）；传 provider 对象——
         // providers[].context 覆盖生效（PROVIDER.md §15 T-C2）
@@ -250,6 +296,10 @@ switch (command) {
         agent.config.agent.compactThreshold = resolveCompactThreshold(null, agent.provider).value
       }
     }
+    // D-3 钉槽：applySession 清 _slot 后立即钉回恢复槽——首保存必落恢复槽（消除
+    // "load → 首回合保存"窗口内并发方翻 active 导致的静默迁移）；/new 经 resetSessionState
+    // 清 _slot（newSession 已认领 + 写记录，语义保留）
+    agent._slot = slot
     // D-S3 优先级补全：applySession 可能已用会话中的有效 provider 修复（config 无效 + 会话有效）——
     // 修复后复验清除标记，仅当两者都无效才弹重选（validateProvider 幂等）
     if (agent._providerInvalid) validateProvider(agent)
@@ -268,7 +318,9 @@ switch (command) {
         projectDir: config.memory.projectDir ? join(process.cwd(), config.memory.projectDir) : null,
         team: teamConfig(config),
         author: gitAuthor(),
-        restored,
+        restored: data,
+        // R25（F-R25c）：TUI 启动显示一行"上次运行异常终止"提示（showStartup 渲染——无匹配不传）
+        crashNotice: recentCrashHint() ?? undefined,
       })
     } catch (error) {
       console.error(`[error] ${error.message}`)
@@ -298,10 +350,10 @@ switch (command) {
         list)   COMPREPLY=( \\$(compgen -W "--type=rule --type=knowledge --type=decision --type=pattern" -- "\\$cur") ) ;;
         put)    COMPREPLY=( \\$(compgen -W "--type= --title= --content= --tags=" -- "\\$cur") ) ;;
       esac ;;
-    distill) COMPREPLY=( \\$(compgen -W "--yes --scope=" -- "\\$cur") ) ;;
+    distill) COMPREPLY=( \\$(compgen -W "--yes --layer=" -- "\\$cur") ) ;;
     completion) COMPREPLY=( \\$(compgen -W "bash zsh fish" -- "\\$cur") ) ;;
     *)
-      COMPREPLY=( \\$(compgen -W "chat acp memory sync reindex distill upgrade completion -v --version -h --help" -- "\\$cur") ) ;;
+      COMPREPLY=( \\$(compgen -W "chat acp memory sync reindex distill upgrade completion session -v --version -h --help" -- "\\$cur") ) ;;
   esac
 }
 complete -F _thincoder thincoder
@@ -326,7 +378,8 @@ _thincoder() {
         'reindex[Rebuild local index from markdown]' \\
         'distill[Extract knowledge from session transcript]' \\
         'upgrade[Update to latest version from npm]' \\
-        'completion[Generate shell completion script]'
+        'completion[Generate shell completion script]' \\
+        'session[Session dir GC: session gc --dry-run|--confirm]'
       ;;
     args)
       case "\\$words[1]" in
@@ -336,7 +389,7 @@ _thincoder() {
             list) _arguments '--type=[Filter by type]' ;;
             put)  _arguments '--type=[Entry type]' '--title=[Title]' '--content=[Content]' '--tags=[Space-separated tags]' ;;
           esac ;;
-        distill) _arguments '--yes[Skip confirmation]' '--scope=[Scope filter]' ;;
+        distill) _arguments '--yes[Skip confirmation]' '--layer=[Layer filter]' ;;
         completion) _values 'shell' 'bash' 'zsh' 'fish' ;;
       esac ;;
   esac
@@ -355,6 +408,7 @@ complete -c thincoder -a reindex  -d 'Rebuild local index from markdown'
 complete -c thincoder -a distill  -d 'Extract knowledge from session'
 complete -c thincoder -a upgrade  -d 'Update to latest version'
 complete -c thincoder -a completion -d 'Shell completion'
+complete -c thincoder -a session -d 'Session dir GC (session gc --dry-run|--confirm)'
 complete -c thincoder -a acp -d 'Agent Client Protocol server for IDEs'
 
 # Flags
@@ -381,7 +435,7 @@ complete -c thincoder -n '__fish_seen_subcommand_from memory; and __fish_seen_su
 
 # distill flags
 complete -c thincoder -n '__fish_seen_subcommand_from distill' -l yes   -d 'Skip confirmation'
-complete -c thincoder -n '__fish_seen_subcommand_from distill' -l scope -d 'Scope filter'
+complete -c thincoder -n '__fish_seen_subcommand_from distill' -l layer -d 'Layer filter'
 
 # completion shells
 complete -c thincoder -n '__fish_seen_subcommand_from completion' -a bash -d 'Bash completions'
@@ -419,6 +473,13 @@ complete -c thincoder -n '__fish_seen_subcommand_from completion' -a fish -d 'Fi
     break
   }
 
+  case "session": {
+    // SESSION.md §12：会话目录 GC 手动面（F2 冷 cwd 报告/删除——VS Code 端无 shell 通道，仅 CLI）
+    const { runSessionGc } = await import("../src/session-gc.mjs")
+    process.exitCode = runSessionGc(args)
+    break
+  }
+
   case "--help":
   case "-h": {
     process.stdout.write(USAGE)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "thincoder",
-  "version": "0.12.59",
+  "version": "0.12.60",
   "description": "Thin coding agent - zero dependencies, no build step, Node.js native. Sharp code, zero bloat.",
   "keywords": [
     "ai",
@@ -31,9 +31,10 @@
     "url": "git+https://github.com/xinbo-tech/thincoder.git"
   },
   "scripts": {
-    "test": "node --test \"test/*.test.mjs\"",
-    "prepublishOnly": "npm run lint && node --test \"test/*.mjs\"",
+    "test": "node test/run-fast.mjs",
+    "prepublishOnly": "npm run release:check",
     "lint": "node scripts/check-syntax.mjs",
+    "release:check": "node scripts/release-check.mjs",
     "test:full": "node test/run-full.mjs"
   },
   "author": "liwei <liwei@51marine.com> (上海新舶)",

package/src/acp/bridge.mjs

@@ -20,7 +20,7 @@
  * `{ stopReason: "end_turn" }` (kimi session.ts parity).
  */
 import { detectDanger, normalizeEOL, joinWithEol } from "../tools/shared.mjs"
-import { computeEditEntry, validateEditEntry, assertEditArgsExclusive } from "../tools/edit-diff.mjs"
+import { computeEditEntry, validateEditEntry, assertEditArgsExclusive, hasLineParams } from "../tools/edit-diff.mjs"
 
 /** ACP ToolKind inference (schema v1 enum) — best-effort, clients render by kind. */
 function inferToolKind(name) {
@@ -92,7 +92,7 @@ export function buildAcpCallbacks({ sessionId, notify, request, log = () => {} }
   }
 
   // §15.1（TOOLS.md）D15.7 委派：edit 判定/应用单一权威 = 本地 computeEditEntry
-  // （edit-diff.mjs——校验→判定序→应用：行级 LCS、零重叠→插入、replace_all 字面替换全部）。
+  // （edit-diff.mjs——校验→判定序→应用：行级 LCS、零重叠→替换即删、replace_all 字面替换全部）。
   // 桥只留「读 IDE 缓冲 → computeEditEntry → 写回 IDE 缓冲」——错误文本经抛错原样透传
   // ——与本地通道逐字一致（NF15.6b / AC15.10：not found / occurrences / 空 old / 空 new）。
   const EDIT_ABORT_PREFIX = "edit aborted (atomic — no files written): "
@@ -130,7 +130,7 @@ export function buildAcpCallbacks({ sessionId, notify, request, log = () => {} }
   const editBatch = async (args) => {
     const edits = args.edits
     if (!Array.isArray(edits) || edits.length === 0) {
-      throw new Error("edits must be a non-empty array of {path, old_string, new_string}")
+      throw new Error("edits must be a non-empty array of {path, old_string | line/startLine+endLine, new_string}")
     }
     assertEditArgsExclusive(args)
     const groups = new Map() // path → { path, raw, content, edits }
@@ -269,7 +269,10 @@ export function buildAcpCallbacks({ sessionId, notify, request, log = () => {} }
           return { handled: true, result: `Error: fs/write_text_file failed: ${e.message}` }
         }
       }
-      if (base === "edit" && (Array.isArray(args?.edits) || (path && typeof args?.old_string === "string" && typeof args?.new_string === "string"))) {
+      // 2026-09-08 D1：单形态按行号改（path + line/startLine/endLine + new_string——无
+      // old_string）同样走 IDE 缓冲通道（hasLineParams 判定）——否则回落本地写盘会与
+      // IDE 缓冲脱敏；editSingle → computeEditEntry 行号语义自动继承。
+      if (base === "edit" && (Array.isArray(args?.edits) || (path && typeof args?.new_string === "string" && (typeof args?.old_string === "string" || hasLineParams(args))))) {
         try {
           const text = Array.isArray(args?.edits) ? await editBatch(args) : await editSingle(path, args)
           return { handled: true, result: text }

package/src/advisor/messages.mjs

@@ -129,6 +129,25 @@ export function buildObjectDeclarationBlock(object = null) {
   ].join("\n")
 }
 
+/**
+ * Approval-signal block for design reviews (round 1 and round 2+ — §24 D-24b:
+ * an async fix-round continuation must be able to re-approve, so the token is
+ * injected into EVERY design round; the reviewer echoes it only on a clean pass).
+ * §29.1 F2a (2026-09-07): BOTH values are injected — the token AND the designId
+ * (anchor sentence verbatim — Copy BOTH values). designId null (legacy direct
+ * callers without a resolved instance) degrades to the token-only form.
+ */
+export function buildDesignApprovalBlock(designToken, designId) {
+  const echo = designId
+    ? `If — and ONLY if — your review finds NO 🔴 (Critical) issues, end your reply with this exact token: [DESIGN-TOKEN:${designToken}] and this exact designId: ${designId}. Copy BOTH values verbatim.`
+    : `If — and ONLY if — your review finds NO 🔴 (Critical) issues, end your reply with this exact token: [DESIGN-TOKEN:${designToken}]`
+  return [
+    "## Approval Signal",
+    echo,
+    "🟡 (Advisory) and 🔵 (Note) findings do NOT block approval — list them if present, but still include the token. If there are any 🔴 issues, do NOT include the token.",
+  ].join("\n")
+}
+
 /**
  * Build the user message for an advisor review session.
  * @param {Object} agent — the parent agent
@@ -142,9 +161,12 @@ export function buildObjectDeclarationBlock(object = null) {
  * @param {Object|null} [object] — review-object declaration (§18.8 D-OA1/D-OA3):
  *   { type, target, status, reason, exclude } — mechanically injected at the
  *   start of the user message; absent → no injection (legacy calls unchanged).
+ * @param {string|null} [designId] — §29.1 F2a: injected next to the token in the
+ *   Approval Signal (both values — the reviewer copies both verbatim); null →
+ *   token-only degradation (legacy direct callers).
  * @returns {string} the user message
  */
-export function buildAdvisorUserMessage(agent, prior, reviewType, designToken = null, documents = null, paths = null, object = null) {
+export function buildAdvisorUserMessage(agent, prior, reviewType, designToken = null, documents = null, paths = null, object = null, designId = null) {
   // prior = the full prior review output (string) when a convergence round is
   // being built (decision 2026-08-08 — verbatim injection, model understands it).
   // Deterministic: only _advisorRound > 0 with stored output counts.
@@ -246,9 +268,7 @@ export function buildAdvisorUserMessage(agent, prior, reviewType, designToken =
     parts.push("5. If you find issues, produce your review table with the format: | # | Category | Severity | Issue | Suggestion |. If the design passes, no table is needed.")
     if (designToken) {
       parts.push("")
-      parts.push("## Approval Signal")
-      parts.push(`If — and ONLY if — your review finds NO 🔴 (Critical) issues, end your reply with this exact token: [DESIGN-TOKEN:${designToken}]`)
-      parts.push("🟡 (Advisory) and 🔵 (Note) findings do NOT block approval — list them if present, but still include the token. If there are any 🔴 issues, do NOT include the token.")
+      parts.push(buildDesignApprovalBlock(designToken, designId))
     }
     return parts.join("\n")
   }

package/src/advisor/run.mjs

@@ -10,11 +10,10 @@ import { appendCitationReport } from "./citations.mjs"
 import { describeToolArgs } from "../tui/tool-args.mjs"
 
 const MAX_ADVISOR_TURNS = 100
-// Mechanical convergence cap: the protocol assumes up to 5 rounds suffice
-// (full review, verify+fix cycles, strict verification). A 6th call means the
-// model is looping — refuse it instead of burning tokens on a review that cannot
-// converge. Code AND design reviews share the 5-round budget (each advances
-// _advisorRound in agent.mjs; the cap no longer exempts design).
+// Mechanical convergence cap: up to 5 rounds suffice; a 6th call means the model
+// is looping — refuse it instead of burning tokens. §24 D-24b (2026-09-06): PER
+// REVIEW INSTANCE (agent._advisorRuns); CODE REVIEWS ONLY (2026-09-07 §8 ruling)
+// — design reviews are EXEMPT: their rounds keep advancing, the cap never refuses.
 export const MAX_ADVISOR_ROUNDS = 5
 
 // NOTE: prompts/advisor-round{1,2,3}.md encourage the model to finish within
@@ -221,7 +220,7 @@ async function runAdvisorToolLoop(provider, messages, onOutput, signal, agent, c
     // already streamed into the timeline via onText; fall back to
     // response.content only if nothing was recorded.
     if (!response.toolCalls?.length) {
-      if (!response.content?.trim()) return renderTimeline(timeline) || "Advisor: (empty response — review was inconclusive)"
+      if (!response.content?.trim()) return renderTimeline(timeline) || "Advisor: empty response — review was inconclusive"
       return renderTimeline(timeline) || response.content.trim()
     }
 
@@ -388,17 +387,35 @@ function extractUnfixedIssues(priorText) {
     .filter(Boolean)
     .slice(0, MAX_UNFIXED_DISPLAY)
 }
+/** Review-looking guard (async settle parity): a markdown table row or ≥200 chars of prose counts as a prior. */
+export function looksLikeReviewOutput(text) {
+  const trimmed = String(text ?? "").trim()
+  return /\|.*\|.*\|/.test(trimmed) || trimmed.length >= 200
+}
+/** Cap message (shared by runAdvisorReview and the async pre-check — per-review refusal). */
+export function buildCapMessage(agent) {
+  const prior = agent._lastAdvisorOutput
+  const unfixed = prior ? extractUnfixedIssues(prior) : []
+  let message = `Advisor: convergence cap reached after ${MAX_ADVISOR_ROUNDS} rounds.\n`
+  if (unfixed.length > 0) {
+    message += `\nUnresolved issues from prior rounds:\n${unfixed.map((i) => `- ${i}`).join("\n")}\n`
+  } else {
+    message += "\nAll prior issues appear resolved.\n"
+  }
+  message += "\nOptions:\n1. Accept current state and proceed\n2. Manually review specific concerns with read/grep\n3. Start a new session (/new) to reset the advisor"
+  return message
+}
 
 /**
  * Run an advisor review. reviewType: "code" (default) or "design". Returns review text or null when skipped.
  * @param {string|null} [designToken] — injected into the design-review prompt; the advisor echoes it only on approval.
  * @param {string[]|null} [documents] — design review only: explicit list of doc paths to review; passed through to the message builder.
  * @param {string[]|null} [paths] — code review only: explicit list of file/dir paths to review.
- * @param {Object|null} [object] — review-object declaration (§18.8 D-OA1/D-OA3):
- *   { type, target, status, reason, exclude }; mechanically injected at the
- *   start of every review round's user message. Absent → legacy behavior (no injection).
+ * @param {Object|null} [object] — review-object declaration (§18.8 D-OA1/D-OA3): { type, target, status, reason, exclude }; absent → legacy behavior (no injection).
+ * @param {string|null} [designId] — §29.1 F2a: injected next to the design token in
+ *   the Approval Signal (passed through to the message builders).
  */
-export async function runAdvisorReview(agent, reviewType, callbacks, designToken = null, documents = null, paths = null, object = null) {
+export async function runAdvisorReview(agent, reviewType, callbacks, designToken = null, documents = null, paths = null, object = null, designId = null) {
   const onOutput = callbacks?.onOutput
   const signal = callbacks?.signal
   const startTime = Date.now()
@@ -407,33 +424,19 @@ export async function runAdvisorReview(agent, reviewType, callbacks, designToken
   // former advisor.enabled gate is removed — review capability has no off
   // switch; only the guard (completion pushback) is opt-in via advisor.guard.
 
-  // Mechanical convergence cap — refuse further reviews once the protocol has run
-  // its rounds. _advisorRound counts completed advisor calls (incremented by the
-  // agent after each one — code AND design reviews alike), so >= MAX_ADVISOR_ROUNDS
-  // blocks the next call. 5 rounds max; after that the review is never pushed back
-  // (the caller decides: accept, manual re-check, or /new to reset).
-  if ((agent._advisorRound || 0) >= MAX_ADVISOR_ROUNDS) {
-    // Summarize unresolved items from the last review output for guidance
-    // (line-level status-word scan — no table-header parsing, decision 2026-08-08).
-    const prior = agent._lastAdvisorOutput
-    const unfixed = prior ? extractUnfixedIssues(prior) : []
-    
-    let message = `Advisor: convergence cap reached after ${MAX_ADVISOR_ROUNDS} rounds.\n`
-    if (unfixed.length > 0) {
-      message += `\nUnresolved issues from prior rounds:\n${unfixed.map((i) => `- ${i}`).join("\n")}\n`
-    } else {
-      message += "\nAll prior issues appear resolved.\n"
-    }
-    message += "\nOptions:\n1. Accept current state and proceed\n2. Manually review specific concerns with read/grep\n3. Start a new session (/new) to reset the advisor"
-    
-    return message
+  // Mechanical convergence cap — CODE REVIEWS ONLY (2026-09-07 §8 ruling: design
+  // reviews are exempt). _advisorRound is scoped to the current review instance
+  // (§24 D-24b ③), so >= MAX_ADVISOR_ROUNDS blocks the next call of THIS instance.
+  // 5 rounds max; after that the review is never pushed back.
+  if (reviewType !== "design" && (agent._advisorRound || 0) >= MAX_ADVISOR_ROUNDS) {
+    return buildCapMessage(agent)
   }
 
   const provider = resolveAdvisorProvider(agent)
   // Advisor always works in the agent's cwd — scope is defined by paths/documents.
   const advisorCwd = agent.cwd
 
-  const messages = prepareAdvisorMessages(agent, reviewType, designToken, documents, paths, null, object)
+  const messages = prepareAdvisorMessages(agent, reviewType, designToken, documents, paths, null, object, designId)
 
   try {
     const result = await runAdvisorToolLoop(provider, messages, onOutput, signal, agent, advisorCwd)
@@ -454,8 +457,7 @@ export async function runAdvisorReview(agent, reviewType, callbacks, designToken
       // table row (`| a | b | c |`) or substantial prose (>200 chars). An
       // empty or tool-progress-only reply must not become the "prior review"
       // of round 2+.
-      const trimmed = final.trim()
-      const looksLikeReview = /\|.*\|.*\|/.test(trimmed) || trimmed.length >= 200
+      const looksLikeReview = looksLikeReviewOutput(final)
       if (looksLikeReview) {
         agent._lastAdvisorOutput = final
       }

package/src/advisor.mjs

@@ -45,13 +45,13 @@ import { readFileSync } from "node:fs"
 import { join, dirname } from "node:path"
 import { fileURLToPath } from "node:url"
 import { extractAgentResponseTable } from "./advisor/history.mjs"
-import { buildAdvisorUserMessage, resolveScopeFiles, buildObjectDeclarationBlock } from "./advisor/messages.mjs"
+import { buildAdvisorUserMessage, resolveScopeFiles, buildObjectDeclarationBlock, buildDesignApprovalBlock } from "./advisor/messages.mjs"
 import { buildConvergenceBody } from "./advisor/convergence.mjs"
 import { escapeLiteralEscapes } from "./escape.mjs"
 // Re-export for run.mjs and tests (keeps their imports from "../advisor.mjs" stable)
 export { ADVISOR_MD_PATH, extractAgentResponseTable, extractConversationBackground } from "./advisor/history.mjs"
 export { buildAdvisorUserMessage } from "./advisor/messages.mjs"
-export { buildObjectDeclarationBlock } from "./advisor/messages.mjs"
+export { buildObjectDeclarationBlock, buildDesignApprovalBlock } from "./advisor/messages.mjs"
 
 const __dirname = dirname(fileURLToPath(import.meta.url))
 
@@ -195,8 +195,10 @@ export { escapeLiteralEscapes }
  * @param {Object|null} [object] — review-object declaration (§18.8 D-OA1): passed through
  *   to the user-message builders; mechanically injected at the start of every review
  *   round (round 1 design/code + round 2+ follow-up). Absent → legacy behavior.
+ * @param {string|null} [designId] — §29.1 F2a: injected next to the design token in
+ *   the Approval Signal (round 1 + round 2+ — both values, verbatim anchor).
  */
-export function prepareAdvisorMessages(agent, reviewType, designToken = null, documents = null, paths = null, priorParam = null, object = null) {
+export function prepareAdvisorMessages(agent, reviewType, designToken = null, documents = null, paths = null, priorParam = null, object = null, designId = null) {
   // Deterministic convergence state (decision 2026-08-08): round 2+ requires
   // _advisorRound > 0 AND a stored prior review output. No history parsing.
   // priorParam (direct callers) wins over the stored output — same derivation
@@ -209,7 +211,7 @@ export function prepareAdvisorMessages(agent, reviewType, designToken = null, do
   if (reviewType === "design" && (agent._advisorRound || 0) === 0) {
     return [
       { role: "system", content: withTime(buildAdvisorSystemPrompt(agent, prior, reviewType)) },
-      { role: "user", content: escapeLiteralEscapes(buildAdvisorUserMessage(agent, prior, reviewType, designToken, documents, paths, object)) },
+      { role: "user", content: escapeLiteralEscapes(buildAdvisorUserMessage(agent, prior, reviewType, designToken, documents, paths, object, designId)) },
     ]
   }
 
@@ -234,7 +236,7 @@ export function prepareAdvisorMessages(agent, reviewType, designToken = null, do
       agent._advisorRound = 0
     }
     // Mutations exist → KEEP the round (cap keeps advancing through retries).
-    const user = buildAdvisorUserMessage(agent, prior, reviewType, designToken, documents, paths, object)
+    const user = buildAdvisorUserMessage(agent, prior, reviewType, designToken, documents, paths, object, designId)
     return [
       { role: "system", content: withTime(buildAdvisorSystemPrompt(agent, prior, reviewType)) },
       {
@@ -263,8 +265,25 @@ export function prepareAdvisorMessages(agent, reviewType, designToken = null, do
   // scopeFiles gives the fallback (agent gave no response table) a concrete
   // review surface.
   const scopeFiles = resolveScopeFiles(agent, paths)
+  const followUp = buildAdvisorFollowUp(agent, prior, scopeFiles, object)
+  // §24 D-24b (design round 2+ — async fix-round continuations must be able to
+  // re-approve): re-anchor the review scope (the convergence follow-up carries no
+  // document list) and inject the round's design token with the approval signal.
+  if (reviewType === "design") {
+    const docList = Array.isArray(documents)
+      ? documents.filter((d) => typeof d === "string" && d.trim())
+      : []
+    const scopeBlock = docList.length > 0
+      ? `\n\n## Documents to Review\nThe documents below are the review scope. Review ONLY these files — do not scan git diff or read any other files.\n${docList.map((d) => `- ${d} — Read this file in full`).join("\n")}`
+      : ""
+    const tokenBlock = designToken ? `\n\n${buildDesignApprovalBlock(designToken, designId)}` : ""
+    return [
+      { role: "system", content: withTime(buildAdvisorSystemPrompt(agent, prior, reviewType)) },
+      { role: "user", content: escapeLiteralEscapes(followUp + scopeBlock + tokenBlock) },
+    ]
+  }
   return [
     { role: "system", content: withTime(buildAdvisorSystemPrompt(agent, prior, reviewType)) },
-    { role: "user", content: escapeLiteralEscapes(buildAdvisorFollowUp(agent, prior, scopeFiles, object)) },
+    { role: "user", content: escapeLiteralEscapes(followUp) },
   ]
 }