AMAP-ML · lerogo · Aug 20, 2026 · Aug 20, 2026
diff --git a/README.md b/README.md
@@ -36,6 +36,7 @@ LongHorizon-Harness turns existing agents into long-running computer-use systems
 
 ## ✨ News
 
+- **[v0.1.7 · 2026-08-20]** A finished run is no longer a dead end: the workbench is now a conversation. Read the reply, type a follow-up, and the run continues on its own round ledger instead of replanning from scratch. A message you send mid-round is claimed by the very next round, so stopping and continuing never drops it. Also adds `--reasoning-effort` for every role (with `--manager-reasoning-effort` and friends to override one), forwarded to whichever backend exposes it. The transcript now reads in strict chronological order, and a graceful stop escalates to a force stop only when a worker ignores it.
 - **[v0.1.6 · 2026-08-15]** Added [OpenCode](https://github.com/anomalyco/opencode) CLI support. LongHorizon-Harness can now run `opencode run prompt` as `--agent opencode`, with role-scoped read/write permissions, OpenCode API endpoint overrides, normalized JSON results, and CLI/config/doctor integration. The Web workbench can select OpenCode Harness and its model independently for each role.
 - **[v0.1.5 · 2026-08-14]** Added phase-1 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) CLI support. LongHorizon-Harness can now run `dsh --profile headless` as `--agent deepseek_harness`, with an isolated `DSH_HOME`, role-scoped read/write permissions, DeepSeek API endpoint overrides, normalized JSONL results, and CLI/config/doctor integration. The Web workbench can select DeepSeek Harness and its model independently for each role. GUI computer-use and MCP support will follow in a later phase; see [the CLI setup](#5-or-run-a-task-from-the-command-line).
 - **[v0.1.4 · 2026-08-11]** The new Dashboard has landed: a React/FastAPI workbench you can drive entirely from the browser. Start a task, choose a backend and model per role, answer approvals, send an instruction mid-run, and stop or restart a run. Launch it with `lh-harness web`; see [Run a task in the browser](#4-run-a-task-in-the-browser-recommended).
@@ -236,7 +237,7 @@ This creates `./.lh-harness/config.toml` without replacing an existing file; use
 lh-harness web --workspace-root .
 ```
 
-This opens the workbench at `http://127.0.0.1:8799/`. Everything happens there: start a task, pick a backend and model per role, answer approval requests, send an instruction mid-run, and stop or restart a run. `--workspace-root` sets the default working directory for tasks created there; the remaining options are listed under [Dashboard commands](#dashboard-commands).
+This opens the workbench at `http://127.0.0.1:8799/`. Everything happens there: start a task, pick a backend and model per role, answer approval requests, send an instruction mid-run, stop or restart a run, and keep asking follow-up questions after it finishes — a follow-up continues the same run from the rounds it already completed. `--workspace-root` sets the default working directory for tasks created there; the remaining options are listed under [Dashboard commands](#dashboard-commands).
 
 #### 5. Or run a task from the command line
 
@@ -327,6 +328,7 @@ Task text, run IDs, and API keys are deliberately **not** configurable here; the
 |---|---|---|
 | `agent` | `"codex"` | Backend for every role unless a role overrides it: `codex`, `claude_code`, `opencode`, or `deepseek_harness`. |
 | `model` | `"gpt-5.6-sol"` | Model for every role unless a role overrides it. Must be a model the chosen backend exposes. |
+| `reasoning_effort` | commented out | Reasoning depth for every role unless a role overrides it, forwarded to whichever backend exposes it. Unset keeps the provider's own setting. |
 | `env` | `"local"` | Execution environment. Only `local` today. |
 | `runs_root` | `"./.lh-harness/runs"` | Where run directories are created. Each run gets `<runs_root>/<run-id>/`. |
 | `workspace` | commented out | Working directory the agents operate in. Defaults to the directory `lh-harness` was started from, so a task acts on your real project; set it to isolate the run somewhere else. |
@@ -354,13 +356,13 @@ Per-episode limits in seconds. One episode is a single role invocation, not the
 
 ##### `[run.roles.*]`
 
-Each role can take its own `agent` and `model`, so you can pay for a strong model only where it matters: a capable Manager and Auditor with a cheaper Executor, for example. Every field is commented out by default, meaning "inherit".
+Each role can take its own `agent`, `model`, and `reasoning_effort`, so you can pay for a strong model only where it matters: a capable Manager and Auditor with a cheaper Executor, for example. Every field is commented out by default, meaning "inherit".
 
 Resolution walks the chain until it finds a value:
 
 ```
-gui_executor → executor → [run].agent / [run].model
-cli_auditor  → auditor  → [run].agent / [run].model
+gui_executor → executor → [run].agent / [run].model / [run].reasoning_effort
+cli_auditor  → auditor  → [run].agent / [run].model / [run].reasoning_effort
 ```
 
 | Section | Falls back to | Covers |

diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -36,6 +36,7 @@ LongHorizon-Harness 将现有 Agent 变成可长期运行的 computer-use 系统
 
 ## ✨ News
 
+- **[v0.1.7 · 2026-08-20]** 任务跑完不再是终点，工作台变成了一场对话：看完回复直接追问，任务会沿用自己已完成的轮次继续跑，而不是从头重新规划。运行中发出的消息会被下一轮立即取用，先停止再继续也不会漏掉。同时新增 `--reasoning-effort` 统一设置各角色的推理强度（也可用 `--manager-reasoning-effort` 等单独覆盖），并转发给支持该能力的后端。对话现在严格按时间顺序展示，而强制中止只在 worker 忽略正常停止时才会出现。
 - **[v0.1.6 · 2026-08-15]** 新增 [OpenCode](https://github.com/anomalyco/opencode) CLI 支持。LongHorizon-Harness 现在可以通过 `--agent opencode` 调用 `opencode run prompt`，并支持按角色划分的读写权限、OpenCode API 端点覆盖、标准化 JSON 结果，以及 CLI/config/doctor 集成。Web 工作台可以为每个角色单独选择 OpenCode Harness 及其模型。
 - **[v0.1.5 · 2026-08-14]** 第一阶段已适配 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) CLI。LongHorizon-Harness 现在可以通过 `--agent deepseek_harness` 调用 `dsh --profile headless`，并提供隔离的 `DSH_HOME`、按角色划分的读写权限、DeepSeek API 端点覆盖、标准化 JSONL 结果以及 CLI/config/doctor 接入。Web 工作台支持为每个角色分别选择 DeepSeek Harness 及其模型；GUI computer-use 和 MCP 支持将在后续阶段补充。使用方式见 [CLI 配置说明](#5-也可以用命令行运行任务)。
 - **[v0.1.4 · 2026-08-11]** 新版 Dashboard 已上线：基于 React/FastAPI 的工作台，全部操作都能在浏览器里完成——发起任务、为每个角色分别选择后端和模型、处理审批、运行中追加指令、停止或重启任务。用 `lh-harness web` 启动，见[在网页上运行任务](#4-在网页上运行任务推荐)。
@@ -237,7 +238,7 @@ lh-harness init
 lh-harness web --workspace-root .
 ```
 
-该命令在 `http://127.0.0.1:8799/` 打开工作台，所有操作都在这里完成：发起任务、为每个角色分别选择后端和模型、处理审批请求、在运行中追加指令、停止或重启任务。`--workspace-root` 指定在工作台中创建任务时的默认工作目录，其余参数见 [Dashboard 命令](#dashboard-命令)。
+该命令在 `http://127.0.0.1:8799/` 打开工作台，所有操作都在这里完成：发起任务、为每个角色分别选择后端和模型、处理审批请求、在运行中追加指令、停止或重启任务，以及在任务结束后继续追问——追问会沿用它已完成的轮次继续同一个任务。`--workspace-root` 指定在工作台中创建任务时的默认工作目录，其余参数见 [Dashboard 命令](#dashboard-命令)。
 
 #### 5. 也可以用命令行运行任务
 
@@ -328,6 +329,7 @@ lh-harness check-update
 |---|---|---|
 | `agent` | `"codex"` | 所有角色使用的后端（角色可单独覆盖）：`codex`、`claude_code`、`opencode` 或 `deepseek_harness`。 |
 | `model` | `"gpt-5.6-sol"` | 所有角色使用的模型（角色可单独覆盖）。必须是所选后端支持的模型。 |
+| `reasoning_effort` | 默认注释 | 所有角色的推理强度（角色可单独覆盖），转发给支持该能力的后端。不设置则沿用服务方自身的默认值。 |
 | `env` | `"local"` | 执行环境，目前只有 `local`。 |
 | `runs_root` | `"./.lh-harness/runs"` | 运行目录的根路径，每次运行生成 `<runs_root>/<run-id>/`。 |
 | `workspace` | 默认注释 | Agent 实际操作的工作目录。默认就是启动 `lh-harness` 的那个目录，任务直接作用于你的真实项目；需要隔离到别处时才设置。 |
@@ -355,13 +357,13 @@ lh-harness check-update
 
 ##### `[run.roles.*]`
 
-每个角色都可以单独指定 `agent` 与 `model`，因此可以只在关键位置使用强模型：例如 Manager 与 Auditor 用强模型、Executor 用更便宜的。所有字段默认都是注释状态，表示「继承」。
+每个角色都可以单独指定 `agent`、`model` 与 `reasoning_effort`，因此可以只在关键位置使用强模型：例如 Manager 与 Auditor 用强模型、Executor 用更便宜的。所有字段默认都是注释状态，表示「继承」。
 
 取值时沿以下链路回退，直到找到值为止：
 
 ```
-gui_executor → executor → [run].agent / [run].model
-cli_auditor  → auditor  → [run].agent / [run].model
+gui_executor → executor → [run].agent / [run].model / [run].reasoning_effort
+cli_auditor  → auditor  → [run].agent / [run].model / [run].reasoning_effort
 ```
 
 | 配置段 | 回退到 | 作用范围 |

diff --git a/frontend/core/src/commands.ts b/frontend/core/src/commands.ts
@@ -1,5 +1,7 @@
 import { MAX_ROUNDS, type Snapshot } from './types';
 
+const STOPPING_STATUSES = new Set(['stopping', 'aborting', 'stop_requested', 'abort_requested']);
+
 export type CommandName =
   | 'help'
   | 'runs'
@@ -33,7 +35,7 @@ export const COMMAND_CATALOG: readonly CommandDefinition[] = [
   { name: 'approve', args: '<approval_id> <action>', description: 'Resolve an approval', capability: 'approvals', requiresRun: true },
   { name: 'stop', description: 'Request a graceful stop', capability: 'stop', requiresRun: true },
   { name: 'abort', description: 'Abort the active run', capability: 'abort', requiresRun: true },
-  { name: 'resume', description: 'Resume a finished run', capability: 'resume', requiresRun: true },
+  { name: 'resume', description: 'Continue a stopped run from its recorded rounds', capability: 'resume', requiresRun: true },
   { name: 'details', description: 'Open run details', requiresRun: true },
   { name: 'events', description: 'Open the event panel', requiresRun: true },
   { name: 'artifacts', description: 'Open the artifact panel', requiresRun: true },
@@ -53,10 +55,19 @@ export interface NewRunOptions {
   workspace?: string;
   maxRounds?: number;
   promptLanguage?: 'en' | 'zh';
-  roles?: Partial<Record<'manager' | 'executor' | 'auditor', { agent?: string; model?: string }>>;
+  roles?: Partial<Record<'manager' | 'executor' | 'auditor', { agent?: string; model?: string; reasoning_effort?: string }>>;
   error?: string;
 }
 
+/**
+ * Effort values reach the backends as inline TOML (Codex) or argv (the others),
+ * so the characters that could terminate a TOML string are rejected here as
+ * well as server-side. An allow-list of tiers would be wrong: the set differs
+ * per backend and per model, and operators must be able to pass a value a newer
+ * backend adds.
+ */
+const EFFORT_RE = /^[A-Za-z0-9._:-]{1,64}$/u;
+
 export function normaliseMaxRounds(value: string | number | null | undefined, fallback = 25): number {
   const safeFallback = Math.min(MAX_ROUNDS, Math.max(1, Number.isSafeInteger(fallback) ? fallback : 25));
   const text = String(value ?? '').trim();
@@ -101,22 +112,32 @@ export function parseNewRunArgs(args: readonly string[]): NewRunOptions {
     if (parseFlags) {
       let handledValueFlag = false;
       for (const flag of [
-        '--agent', '--model', '--workspace',
-        '--manager-agent', '--manager-model',
-        '--executor-agent', '--executor-model',
-        '--auditor-agent', '--auditor-model',
+        '--agent', '--model', '--workspace', '--effort',
+        '--manager-agent', '--manager-model', '--manager-effort',
+        '--executor-agent', '--executor-model', '--executor-effort',
+        '--auditor-agent', '--auditor-model', '--auditor-effort',
       ] as const) {
         const parsed = valueFor(token, flag, args[index + 1]);
         if (parsed.error) return { task: task.join(' ').trim(), error: parsed.error };
         if (parsed.value !== undefined) {
+          if (flag === '--effort' || flag.endsWith('-effort')) {
+            if (!EFFORT_RE.test(parsed.value)) {
+              return { task: task.join(' ').trim(), error: `${flag} 只能包含字母、数字、'.'、'_'、':' 或 '-'，且不超过 64 个字符` };
+            }
+          }
           if (flag === '--agent') agent = parsed.value;
           else if (flag === '--model') model = parsed.value;
           else if (flag === '--workspace') workspace = parsed.value;
-          else {
-            const match = /^--(manager|executor|auditor)-(agent|model)$/u.exec(flag);
+          else if (flag === '--effort') {
+            // A global --effort applies to every role, matching --agent/--model.
+            for (const role of ['manager', 'executor', 'auditor'] as const) {
+              roles[role] = { ...(roles[role] || {}), reasoning_effort: parsed.value };
+            }
+          } else {
+            const match = /^--(manager|executor|auditor)-(agent|model|effort)$/u.exec(flag);
             if (match) {
               const role = match[1] as 'manager' | 'executor' | 'auditor';
-              const field = match[2] as 'agent' | 'model';
+              const field = match[2] === 'effort' ? 'reasoning_effort' : (match[2] as 'agent' | 'model');
               roles[role] = { ...(roles[role] || {}), [field]: parsed.value };
             }
           }
@@ -194,7 +215,10 @@ export function availableCommands(
     if (command.capability && capabilities[command.capability] === false) return false;
     if (command.name === 'inject' && snapshot && !snapshot.controls.can_inject) return false;
     if (command.name === 'stop' && snapshot && !snapshot.controls.can_abort) return false;
-    if (command.name === 'abort' && snapshot && !snapshot.controls.can_abort) return false;
+    // `/abort` stays typable while stopping: escalating an ignored SIGTERM is
+    // the only lifecycle transition left, and `can_abort` is false by then.
+    if (command.name === 'abort' && snapshot && !snapshot.controls.can_abort
+      && !STOPPING_STATUSES.has(String(snapshot.run.status || '').trim().toLowerCase())) return false;
     if (command.name === 'resume' && snapshot && !snapshot.controls.can_resume) return false;
     return true;
   });