duaer-spec 0.14.0 → 0.16.0

package/.duaer/active-job.json

@@ -1,6 +1,6 @@
 {
-  "featureDir": ".duaer/specs/053-release-014",
-  "branch": "feat/release-014",
-  "worktree": ".worktree/feat-release-014",
-  "status": "active"
+  "featureDir": ".duaer/specs/060-release-016",
+  "branch": "feat/release-016",
+  "worktree": ".worktree/feat-release-016",
+  "status": "accepted"
 }

package/.duaer/memory/testing.md

@@ -1,21 +1,22 @@
 # Testing expectations (project contract)
 
 Agents **must** read this file before marking work verified or accepting a job.
-Replace command examples with the adopting project's real scripts. Keep the
-level model and risk rules unless an ADR documents a change.
+This file is the **duaer-spec** repo contract. Adopters replace command examples
+with their product scripts; keep the level model and risk rules unless an ADR
+documents a change.
 
 ## Levels
 
-| Level | What | Typical commands (replace) | Default owner |
+| Level | What | Typical commands (this repo) | Default owner |
 |---|---|---|---|
-| **L0** | Lint / typecheck / static | `npm run lint`, `tsc --noEmit` | After every implementation pass |
-| **L1** | Unit tests for changed modules | `npm test` (or package subset) | Default for code changes |
+| **L0** | Lint / typecheck / static | `node --check bin/duaer-live.mjs` `node --check web/live-dev/app.js` | After every implementation pass |
+| **L1** | Unit tests for changed modules | `npm test` when unit files exist under `test/` | Default for code changes |
 | **L2** | Integration / product acceptance at changed boundaries | project acceptance or integration suite | When boundaries or core paths change |
-| **L3** | Automated E2E / browser for user-visible paths | `npm run test:e2e` or equivalent **subset** | When UI / protocol UX changes |
+| **L3** | Automated E2E / protocol smoke for user-visible paths | **`npm run test:live`** (live desk shell + `/api/validate` gate; mock LLM, no paid API) | When live desk UI / protocol UX changes |
 | **L4** | Manual account / device path when automation cannot cover auth or real data | documented steps in the Brief | When UI/auth and no automated substitute |
 
 IDE Browser / MCP exploration is **optional discovery**. It does **not** replace
-L3 automation when the project has an E2E suite.
+L3 when `npm run test:live` covers the touched path.
 
 ## Risk table (what to run)
 
@@ -24,12 +25,30 @@ L3 automation when the project has an E2E suite.
 | Docs / comments only | None (note in Brief) |
 | Internal code, same behavior | L0 + L1 for affected packages |
 | API / data / core path | L0 + L1 + L2 (or documented equivalent) |
-| User-visible or protocol-visible UI/UX | L0 + L1 + **L3 subset for touched flows** + update E2E catalog |
+| User-visible or protocol-visible UI/UX (live desk) | L0 + **`npm run test:live` (L3)** + update E2E catalog |
 | Auth / payments / irreversible ops | Above + L4 if L3 cannot cover |
 
 Prefer **targeted subsets** over full monorepo suites. Do not skip a required
 level without an explicit waiver in the feature docs (`tasks.md` or Spec).
 
+## Live desk L3 (`npm run test:live`)
+
+Covers without a paid LLM:
+
+1. Desk shell HTML/CSS/JS markers (three columns, structured confirm fields,
+   `validationAllowsSend`)
+2. Local validate failures (empty / too-short card)
+3. `/api/validate` pass and `/api/validate/fix` via an in-process mock
+   OpenAI-compatible server
+
+Also run `npm test` (includes revise-reliability contract tests: Brief
+snapshot/rollback helpers + source markers for `PREEMPT_FAILED` / longer
+client timeout).
+
+Does **not** cover: real model quality, live Terminal CLI launch against a
+real Cursor/Claude session. Those stay manual E2E catalog rows (E2E-055) or
+future suites.
+
 ## Opt-in (still need an explicit ask or Spec note)
 
 Mark these in the project table when they are expensive:

package/.duaer/specs/054-live-l3-smoke/delivery.json

@@ -0,0 +1,9 @@
+{
+  "status": "accepted",
+  "summary": "Added npm run test:live L3 smoke for desk shell + validate gate; documented in manuals and testing.md; CI on develop/main.",
+  "verification": {
+    "levels": ["L0", "L3"],
+    "commands": ["npm run test:live", "node --check bin/duaer-live.mjs"],
+    "notes": "Mock LLM; no paid API. E2E-054."
+  }
+}

package/.duaer/specs/054-live-l3-smoke/spec.md

@@ -0,0 +1,41 @@
+# Feature Specification: Live desk L3 smoke suite
+
+**Feature Branch**: `feat/live-l3-smoke`
+
+**Created**: 2026-09-17
+
+**Status**: Active
+
+## Goal
+
+Add an automated L3 smoke suite for the live desk (confirm validate gate +
+desk shell markers) that agents run before accept on live UI changes, and
+document it in the manuals / testing contract.
+
+## In scope
+
+- `npm run test:live` — start isolated live + mock LLM; cover validate fail/pass,
+  validate/fix path, static desk markers (three columns, structured fields,
+  validationAllowsSend)
+- Wire into `.duaer/memory/testing.md` as the L3 command for this repo
+- Update README EN/ZH + CHANGELOG + E2E catalog
+- Optional lightweight CI job on push/PR
+
+## Out of scope
+
+- Full Playwright browser interaction / paid live LLM
+- Revise Terminal dispatch end-to-end
+- npm version bump / publish (docs only unless user asks)
+
+## Acceptance
+
+1. `npm run test:live` passes without network LLM
+2. Covers: empty/short card fail, good card pass via mock, fix endpoint,
+   HTML/JS/CSS desk markers for layout + gate
+3. README (EN/ZH) and testing.md document the command
+4. E2E catalog notes automated subset
+
+## Assumptions
+
+- Worktree: `.worktree/feat-live-l3-smoke`
+- Brief: `.duaer/specs/054-live-l3-smoke/`

package/.duaer/specs/054-live-l3-smoke/tasks.md

@@ -0,0 +1,10 @@
+# Tasks: Live desk L3 smoke suite
+
+**Brief**: `.duaer/specs/054-live-l3-smoke/`
+
+## Checklist
+
+- [x] Add `test/live-smoke.test.mjs` + mock LLM helper
+- [x] `package.json` script `test:live`; update testing.md
+- [x] README EN/ZH + CHANGELOG + E2E-054; CI workflow
+- [x] Run `npm run test:live`; accept; merge develop

package/.duaer/specs/055-release-015/delivery.json

@@ -0,0 +1,8 @@
+{
+  "status": "accepted",
+  "summary": "Release 0.15.0 with live L3 smoke suite.",
+  "verification": {
+    "levels": ["L3"],
+    "commands": ["npm run test:live"]
+  }
+}

package/.duaer/specs/055-release-015/spec.md

@@ -0,0 +1,16 @@
+# Feature Specification: Release 0.15.0
+
+**Feature Branch**: `feat/release-015`
+
+**Created**: 2026-09-17
+
+**Status**: Active
+
+## Goal
+
+Ship **0.15.0** (live L3 smoke) to main + npm.
+
+## Acceptance
+
+1. package.json 0.15.0; CHANGELOG section dated
+2. main tagged v0.15.0; GitHub Release; npm latest 0.15.0

package/.duaer/specs/055-release-015/tasks.md

@@ -0,0 +1,4 @@
+# Tasks: Release 0.15.0
+
+- [x] Bump 0.15.0 + CHANGELOG
+- [ ] Promote main; push; tag; gh release

package/.duaer/specs/056-revise-reliability/delivery.json

@@ -0,0 +1,13 @@
+{
+  "status": "accepted",
+  "summary": "Revise dispatch: honest preempt, Brief rollback on launch fail, structured retryable errors, terminal status line, longer client wait.",
+  "verification": {
+    "levels": ["L0", "L3"],
+    "commands": [
+      "node --check bin/duaer-live.mjs",
+      "npm run test:live",
+      "node --test test/revise-reliability.test.mjs"
+    ],
+    "notes": "E2E-055 manual for live Terminal preempt path."
+  }
+}

package/.duaer/specs/056-revise-reliability/spec.md

@@ -0,0 +1,44 @@
+# Feature Specification: Revise round reliability
+
+**Feature Branch**: `feat/revise-reliability`
+
+**Created**: 2026-09-17
+
+**Status**: Active
+
+## Goal
+
+Revise dispatch is **observable**, **retryable**, and failures are **readable**:
+confirm → Terminal enqueue must not report success when preempt failed; Brief
+mutations must not stick after a failed launch; client timeout matches server
+budgets; status exposes Terminal busy/queue; UI keeps dispatch unlocked when
+retryable.
+
+## In scope
+
+- Honest preempt outcome (no false preempted / false HTTP 200)
+- Rollback Brief writes if launch/enqueue fails
+- Structured revise errors (`code`, `retryable`) + UI copy
+- Align client `/api/revise` abort with recreate/init budgets (~3 min)
+- `/api/status` includes `terminal: { busy, queueDepth, runnerHealthy }`
+- Minimal status/hint line; extend L3 smoke for error shape / rollback unit
+
+## Out of scope
+
+- Desk layout / column redesign
+- Full Playwright Terminal automation
+- Paid live LLM quality
+
+## Acceptance
+
+1. When preempt cannot clear leftover agents, `/api/revise` fails with
+   `PREEMPT_FAILED` and does **not** lock the revise card
+2. Failed launch does not leave a new Revision block in Brief (rollback)
+3. Client wait ≥ worst-case recreate path; timeout message mentions retry
+4. Status poll shows terminal busy/queue when revising
+5. `npm run test:live` still passes; new smoke/unit covers rollback or error codes
+
+## Assumptions
+
+- Worktree: `.worktree/feat-revise-reliability`
+- Brief: `.duaer/specs/056-revise-reliability/`

package/.duaer/specs/056-revise-reliability/tasks.md

@@ -0,0 +1,10 @@
+# Tasks: Revise round reliability
+
+**Brief**: `.duaer/specs/056-revise-reliability/`
+
+## Checklist
+
+- [x] Honest preempt + PREEMPT_FAILED; Brief rollback on launch fail
+- [x] Structured errors on `/api/revise`; status.terminal
+- [x] UI: longer timeout, retryable unlock, terminal line, i18n
+- [x] L3 / unit smoke; docs; accept; merge develop

package/.duaer/specs/057-progress-accepted/delivery.json

@@ -0,0 +1,9 @@
+{
+  "status": "accepted",
+  "summary": "On delivery accepted, progress shows N/N and reconciles unchecked tasks.md boxes.",
+  "verification": {
+    "levels": ["L0", "L3"],
+    "commands": ["npm test"],
+    "notes": "E2E-056."
+  }
+}

package/.duaer/specs/057-progress-accepted/spec.md

@@ -0,0 +1,42 @@
+# Feature Specification: Progress sync when delivery accepted
+
+**Feature Branch**: `fix/progress-accepted`
+
+**Created**: 2026-09-17
+
+**Status**: Active
+
+## Goal
+
+Task progress must not show `0/N · delivery accepted · 工单完成`. When
+`delivery.json` is `accepted`, the desk treats tasks as complete (done ===
+total) and reconciles unchecked boxes in `tasks.md` when safe.
+
+## Problem
+
+`dispatchStatus` only rewrites `progress.current` on accept; `done` stays at
+the unchecked count. Agents often stamp `delivery.json` without flipping every
+`- [ ]` to `- [x]`, so 任务跟进 shows a contradictory `0/3 · …工单完成`.
+
+## In scope
+
+- Normalize progress when accepted: `done = total`, all task rows `done: true`
+- Reconcile `tasks.md` unchecked boxes to `[x]` once when accepted (idempotent)
+- Unit/smoke coverage; short README/CHANGELOG/E2E note
+
+## Out of scope
+
+- Changing how agents are prompted (except if a one-line reminder helps)
+- Progress column layout redesign
+
+## Acceptance
+
+1. Accepted delivery never surfaces `done < total` with an “工单完成 /
+   delivery accepted” current line
+2. Unchecked tasks in Brief are marked `[x]` after status observes accept
+3. `npm test` / `npm run test:live` pass
+
+## Assumptions
+
+- Worktree: `.worktree/fix-progress-accepted`
+- Brief: `.duaer/specs/057-progress-accepted/`

package/.duaer/specs/057-progress-accepted/tasks.md

@@ -0,0 +1,8 @@
+# Tasks: Progress sync when delivery accepted
+
+**Brief**: `.duaer/specs/057-progress-accepted/`
+
+## Checklist
+
+- [x] Normalize progress on accept; reconcile tasks.md
+- [x] Tests + docs; accept; merge develop

package/.duaer/specs/058-revise-progress-track/delivery.json

@@ -0,0 +1,13 @@
+{
+  "status": "accepted",
+  "acceptedAt": "2026-09-17T09:50:00.000Z",
+  "summary": "Revise-round progress scopes to Revision N / R{n}-* tasks; lagging accepted + busy Terminal no longer freezes on 工单完成.",
+  "verification": {
+    "levels": ["L0", "L1", "L3"],
+    "commands": [
+      "node --check bin/duaer-live.mjs",
+      "npm test"
+    ],
+    "result": "pass (14 tests)"
+  }
+}

package/.duaer/specs/058-revise-progress-track/spec.md

@@ -0,0 +1,48 @@
+# Feature Specification: Track revise-round task progress
+
+**Feature Branch**: `fix/revise-progress-track`
+
+**Created**: 2026-09-17
+
+**Status**: Accepted
+
+## Goal
+
+After **Confirm revise and dispatch**, 任务跟进 shows **this revision’s**
+checklist and updates as work starts — not the prior dispatch’s completed
+`T001…` pile, and not a frozen “delivery accepted · 工单完成” while Terminal
+is already running the revise.
+
+## Problem
+
+Progress parses the whole `tasks.md`. After revise, old tasks stay `[x]` so
+the bar looks finished (or accepted-normalized to N/N) while the new `R{n}-*`
+round has started. Live `revisionCount` can lag; status may still say
+`accepted`, so the wrong run-block is updated and revise work looks untracked.
+
+## In scope
+
+- Scope progress to dispatch (pre-Revision section) vs active `Revision N`
+- Infer active revision from job / delivery / tasks.md headers and `R{n}` ids
+- While revising (or Terminal busy with open `R{n}` tasks), do not force
+  accept-complete progress
+- Persist inferred `revisionCount` onto the live job when missing
+- Tests + CHANGELOG / E2E note
+
+## Out of scope
+
+- Layout / column redesign
+- Changing revise launch / preempt behavior
+
+## Acceptance
+
+1. After Confirm revise, progress shows only this revision’s tasks (e.g. `0/3`
+   then rising), under the Revision N run block
+2. While revise is active, status is not stuck on `delivery accepted · 工单完成`
+3. When revise delivery accepts, that revision shows `N/N · …完成`
+4. `npm test` passes
+
+## Assumptions
+
+- Worktree: `.worktree/fix-revise-progress-track`
+- Brief: `.duaer/specs/058-revise-progress-track/`

package/.duaer/specs/058-revise-progress-track/tasks.md

@@ -0,0 +1,9 @@
+# Tasks: Track revise-round task progress
+
+**Brief**: `.duaer/specs/058-revise-progress-track/`
+
+## Checklist
+
+- [x] Scope parseTasksProgress by revision; infer active revision
+- [x] Revising / busy Terminal skips accept-complete normalize
+- [x] Tests + docs; merge develop

package/.duaer/specs/059-revise-preempt-order/delivery.json

@@ -0,0 +1,13 @@
+{
+  "status": "accepted",
+  "acceptedAt": "2026-09-17T10:05:00.000Z",
+  "summary": "priorAccepted revise preempts leftovers before enqueue so Confirm revise no longer returns PREEMPT_FAILED while the revise Agent is already running.",
+  "verification": {
+    "levels": ["L0", "L1", "L3"],
+    "commands": [
+      "node --check bin/duaer-live.mjs",
+      "npm test"
+    ],
+    "result": "pass (14 tests)"
+  }
+}

package/.duaer/specs/059-revise-preempt-order/spec.md

@@ -0,0 +1,47 @@
+# Feature Specification: Honest revise preempt (enqueue after clear)
+
+**Feature Branch**: `fix/revise-preempt-order`
+
+**Created**: 2026-09-17
+
+**Status**: Accepted
+
+## Goal
+
+After Confirm revise, either the revise Agent is launched and the desk
+reports success, or preempt truly failed and nothing was started — never
+“Could not preempt…” while Terminal is already running the revise job.
+
+## Problem
+
+`launchInTerminal` enqueues the revise job **before** preempt. Within the
+post-preempt wait, the runner starts the new Agent; status then sees
+`busy && agentsLeft > 0` and throws `PREEMPT_FAILED`. The API rolls back
+Brief while the queued revise Agent keeps working — UI error, work already
+underway; retries kill that Agent.
+
+## In scope
+
+- Preempt leftover agents **before** enqueueing the revise Terminal job
+- Fail `PREEMPT_FAILED` only when prior agents remain (nothing queued yet)
+- Do not treat the newly started revise Agent as a preempt failure
+- Tests + CHANGELOG / E2E note
+
+## Out of scope
+
+- Progress scoping (058)
+- Changing when priorAccepted revise preempts vs waits
+
+## Acceptance
+
+1. priorAccepted revise: preempt clears leftovers, then enqueue; HTTP 200
+2. If leftover agents cannot be cleared, `PREEMPT_FAILED` and **no** new
+   Terminal job for that attempt
+3. Desk does not show preempt failure while revise Agent is already running
+   from that same Confirm tap
+4. `npm test` passes
+
+## Assumptions
+
+- Worktree: `.worktree/fix-revise-preempt-order`
+- Brief: `.duaer/specs/059-revise-preempt-order/`

package/.duaer/specs/059-revise-preempt-order/tasks.md

@@ -0,0 +1,8 @@
+# Tasks: Honest revise preempt (enqueue after clear)
+
+**Brief**: `.duaer/specs/059-revise-preempt-order/`
+
+## Checklist
+
+- [x] Preempt before enqueue; fail only on leftover prior agents
+- [x] Tests + docs; merge develop

package/.duaer/specs/060-release-016/delivery.json

@@ -0,0 +1,10 @@
+{
+  "status": "accepted",
+  "acceptedAt": "2026-09-18T00:00:00.000Z",
+  "summary": "Release 0.16.0: honest revise preempt, revise progress scope, accept progress sync.",
+  "verification": {
+    "levels": ["L0", "L1", "L3"],
+    "commands": ["npm test"],
+    "result": "pass (14 tests)"
+  }
+}

package/.duaer/specs/060-release-016/spec.md

@@ -0,0 +1,24 @@
+# Feature Specification: Release 0.16.0
+
+**Feature Branch**: `feat/release-016`
+
+**Created**: 2026-09-18
+
+**Status**: Accepted
+
+## Goal
+
+Ship **0.16.0** to `main` + npm: revise reliability (honest preempt order),
+revise-round progress tracking, and accept progress sync.
+
+## Acceptance
+
+1. `package.json` version `0.16.0`; CHANGELOG dated section; Unreleased cleared
+2. `main` tagged `v0.16.0`; GitHub Release; npm `latest` is `0.16.0`
+3. `npm test` passes before promote
+
+## Assumptions
+
+- Worktree: `.worktree/feat-release-016`
+- Brief: `.duaer/specs/060-release-016/`
+- User ask「发版」authorizes push + promote `develop` → `main`

package/.duaer/specs/060-release-016/tasks.md

@@ -0,0 +1,4 @@
+# Tasks: Release 0.16.0
+
+- [x] Bump 0.16.0 + CHANGELOG + light README notes
+- [x] Verify `npm test`; merge develop; promote main; tag + npm

package/CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## Unreleased
+
+## 0.16.0 — 2026-09-18
+
+### Honest revise preempt
+
+- priorAccepted revise preempts leftover Agents **before** enqueueing the new
+  Terminal job (no false `PREEMPT_FAILED` while the revise Agent already runs)
+- `PREEMPT_FAILED` only when leftovers remain — nothing queued for that attempt
+
+### Revise progress tracking
+
+- After Confirm revise, task progress scopes to the active `Revision N` checklist
+  (not the prior dispatch’s completed `T*` pile)
+- While revise is active (or Terminal is busy on open `R{n}` work), status does
+  not freeze on the accepted “job complete” label
+- Live `revisionCount` is inferred from Brief/`tasks.md` when the job stamp lags
+
+### Progress sync when delivery accepted
+
+- When `delivery.json` is `accepted`, task progress shows `N/N` (not `0/N` with
+  an accepted label)
+- Unchecked `tasks.md` boxes are reconciled to `[x]` on status poll after accept
+
+### Revise round reliability
+
+- Honest Terminal preempt: no false `preempted` / HTTP 200 when agents still run (`PREEMPT_FAILED`, retryable)
+- Rollback Brief `spec.md` / `tasks.md` / `delivery.json` if revise launch fails
+- `/api/status` includes `terminal: { busy, queueDepth, runnerHealthy }`; desk shows the line
+- Client revise wait raised to ~3 minutes; structured error codes + keep Confirm revise unlocked on failure
+- Stuck revising (busy Terminal, no task progress) resurfaces **Revise again**
+
+## 0.15.0 — 2026-09-17
+
+### Live desk L3 smoke
+
+- `npm run test:live` — automated desk shell + `/api/validate` / `/api/validate/fix` with mock LLM
+- Documented in `.duaer/memory/testing.md`, README EN/ZH; CI runs on push/PR to develop/main
+
 ## 0.14.0 — 2026-09-17
 
 ### Live desk layout

package/README.md

@@ -64,6 +64,10 @@ Open the printed URL (default `http://127.0.0.1:8787`). The desk is **full-width
 with three columns: **chat** | **requirements / confirm / dispatch / revise** |
 **task progress**.
 
+**Verify (maintainers / agents):** after live desk UI or validate-gate changes,
+run `npm run test:live` (L3 smoke: desk shell markers + `/api/validate` with a
+mock LLM — no paid API).
+
 **Flow**
 
 1. Dialogue → fill the confirm card (structured goal / out-of-scope / acceptance /
@@ -73,7 +77,7 @@ with three columns: **chat** | **requirements / confirm / dispatch / revise** |
 3. Choose a **CLI** digital employee: **Cursor Agent** or **Claude Code** (install Cursor CLI if needed: `curl https://cursor.com/install -fsS | bash`)  
 4. Edit the start command (must begin with `Duaer`) → dispatch creates `.worktree/feat-*` and opens **Terminal** to run the CLI  
 5. Watch progress in the right column (`tasks.md`); when `delivery.json` is `accepted`, open **View product** (`preview.url` or auto-detected `index.html`)  
-6. If the result is not right: click **Continue improving (left chat)**, clarify why and what to change, confirm the revise card (same validate gate), then the desk relaunches with `agent --continue` / `claude --continue` on the same worktree  
+6. If the result is not right: click **Continue improving (left chat)**, clarify why and what to change, confirm the revise card (same validate gate), then the desk relaunches on the same worktree (preempt leftover Agents before enqueueing revise). Progress tracks this revision’s `R{n}-*` tasks. Failed launch rolls back the Brief Revision and keeps **Confirm revise** available; status shows Terminal busy/queue.  
 7. If the job needs hosting / a public URL, deploy with **GitHub CLI (`gh`) + Actions** (Pages template under `.duaer/templates/deploy-github-pages.yml`; see [`docs/agent/deploy-github.md`](docs/agent/deploy-github.md))
 
 Worktree id ≠ Brief folder name — use `.worktree/feat-<name>/`, Brief under `.duaer/specs/<nnn-slug>/` inside that worktree.

package/README.zh-CN.md

@@ -64,6 +64,10 @@ duaer live
 浏览器打开提示地址（默认 `http://127.0.0.1:8787`）。台面**全屏宽**，三列：
 **对话** | **需求 / 确认 / 派工 / 改进** | **任务进度**。
 
+**验收（维护者 / 数字员工）：** 改现场台 UI 或确认校验门禁后，跑
+`npm run test:live`（L3 冒烟：台面结构标记 + `/api/validate`，内置 mock 模型，
+不产生付费 API 调用）。
+
 **流程**
 
 1. 多轮对话 → 确认卡（目标 / 不做 / 验收 / 假设，结构化展示）。须**校验通过**后才能点确认；失败可自动修订  
@@ -71,7 +75,7 @@ duaer live
 3. 只选 **CLI 数字员工**：**Cursor Agent** 或 **Claude Code**（未装 Cursor 时页面会提示：`curl https://cursor.com/install -fsS | bash`）  
 4. 编辑「启动命令」（须以 `Duaer` 开头）→ 派工建 `.worktree/feat-*`，并打开 **Terminal** 跑对应 CLI  
 5. 右侧进度列轮询 `tasks.md`；`delivery.json` 为 `accepted` 后出现 **查看成品**（`preview.url`，或自动发现 `index.html`）  
-6. 成品不满意：点 **继续改进（左侧对话）**，说清原因与期望，确认改进卡（同样须校验通过）后，用 Terminal `agent --continue` / `claude --continue` 续派（`POST /api/revise`）  
+6. 成品不满意：点 **继续改进（左侧对话）**，说清原因与期望，确认改进卡（同样须校验通过）后续派（先清残留 Agent 再入队）。进度跟本轮 `R{n}-*` 任务。启动失败会回滚本轮 Brief Revision，并可再点「再派一版」；状态行显示 Terminal 忙闲/队列。  
 7. 若需求需要上线 / 公网访问：默认用 **GitHub CLI（`gh`）+ Actions** 自动化部署（静态站可参考 `.duaer/templates/deploy-github-pages.yml`；说明见英文 [`docs/agent/deploy-github.md`](docs/agent/deploy-github.md)）
 
 ### 文档语言