@tea-agent/loop-agent 0.6.0 → 0.7.0

package/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 这里记录每个版本面向使用者的主要变化。保持简短即可：新增、修改、修复或删除了什么，不需要展开实现细节。
 
+## [Unreleased]
+
+## [0.7.0] - 2026-07-11
+
+### 新增
+
+- TaskSpec 新增可选 `capabilities: [interactive-ui]`：在不改变 `risk_level` / governance profile 的前提下，仅将 implement/repair writer 路由到 HIGH canonical model，并注入真实 UI component、integration 与 DOM interaction test 交付契约。
+- Round-2 FE dogfood 新增真实 React/ReactDOM + Testing Library/user-event/jsdom fixture、AC-FE-001 DOM 验收，以及 prompt-only MED / default-prompt HIGH / UI-prompt HIGH 的 A/B/C DAG 变体生成脚本与实验记录模板。
+- dogfood Feature `F-2026-001` 新增第二个前端任务 `FE-002`（frontend session store，独立于 FE-001）与对应验收项 `AC-FE-002`，补齐“第二个 FE”覆盖。
+
+### 修复
+
+- 修复 `scripts-local/setup-drill-round1.sh` 在 `.task-pool/` 已忽略时仍尝试提交 state、覆盖 init 生成的 `.gitignore` managed block、依赖版本漂移及危险目标路径缺少保护的问题。
+- 新增首轮夜间演练报告 `docs/reports/2026-07-10-nightly-drill-round-1.md`：记录 0.6.0 已发布控制器下的 batch/晨报/决策闭环、review-gated 失败与人工接管证据。
+- 新增可复现演练目标仓库 setup 脚本 `scripts-local/setup-drill-round1.sh`。
+
 ## [0.6.0] - 2026-07-10
 
 ### 新增

package/dist/task/config-types.js

@@ -19,6 +19,7 @@ export const referenceDocConfigSchema = z.object({
     path: z.string().min(1),
 });
 export const taskComplexitySchema = z.enum(["small", "medium", "large"]);
+export const taskCapabilitySchema = z.enum(["interactive-ui"]);
 export const contextProfileSchema = z.enum(["full", "slim"]);
 export const piSubagentModeSchema = z.enum(["off", "analyze-plan", "full"]);
 export const verifyModeSchema = z.enum(["parallel", "serial"]);
@@ -68,6 +69,8 @@ export const taskConfigSchema = z.object({
     timeoutMs: z.number().int().positive().optional(),
     flow: taskFlowSchema.optional().default("auto"),
     complexity: taskComplexitySchema.optional().default("medium"),
+    /** Explicit delivery capabilities that affect writer contracts/model tier without changing risk. */
+    capabilities: z.array(taskCapabilitySchema).optional(),
     verifyMode: verifyModeSchema.optional().default("parallel"),
     verifyPreset: verifyPresetSchema.optional().default("auto"),
     /** Verification selection policy. Intermediate loops may use quota; final gates still run full required verification. */

package/dist/worker/materialize/harness-task-materializer.js

@@ -40,6 +40,7 @@ export async function materializeTaskSpec(options) {
         forbiddenPaths: options.taskSpec.constraints.forbidden_paths,
         hardConstraints: options.taskSpec.constraints.hard_constraints,
         complexity: mapRiskLevelToComplexity(options.taskSpec.risk_level),
+        capabilities: options.taskSpec.capabilities ?? [],
         verifyMode: options.taskSpec.verify.mode,
         verifyPreset: options.taskSpec.verify.preset,
         verifyQuota: options.taskSpec.verify.quota,
@@ -135,6 +136,9 @@ export function renderRequirementMarkdown(taskSpec, options = {}) {
         `Risk level: ${taskSpec.risk_level}`,
         "",
     ];
+    if ((taskSpec.capabilities?.length ?? 0) > 0) {
+        lines.push("## Capabilities", "", ...bulletLines(taskSpec.capabilities ?? []), "");
+    }
     if (taskSpec.description) {
         lines.push("## Description", "", taskSpec.description, "");
     }

package/dist/worker/task-spec/schema.js

@@ -1,5 +1,5 @@
 import { z } from "zod";
-import { verifyModeSchema, verifyPresetSchema, verifyQuotaSchema, } from "../../task/config-types.js";
+import { taskCapabilitySchema, verifyModeSchema, verifyPresetSchema, verifyQuotaSchema, } from "../../task/config-types.js";
 export const taskSpecTypeSchema = z.enum([
     "architecture",
     "backend-feature",
@@ -42,6 +42,7 @@ export const taskSpecSchema = z
     type: taskSpecTypeSchema,
     priority: z.enum(["P0", "P1", "P2", "P3"]).optional().default("P2"),
     risk_level: taskSpecRiskLevelSchema,
+    capabilities: z.array(taskCapabilitySchema).optional(),
     depends_on: z.array(z.string().min(1)).optional().default([]),
     source_docs: z
         .object({

package/dist/workflows/dag/init-hybrid.js

@@ -17,6 +17,32 @@ const CONSTRAINT_FILE = "执行约束.md";
 const REFERENCE_DIRECTORY = "references";
 const MAX_SOURCE_EXCERPT_CHARS = 2000;
 const MAX_SOURCE_REFERENCE_DOCUMENTS = 8;
+const INTERACTIVE_UI_DELIVERY_CONTRACT = [
+    "[INTERACTIVE_UI_DELIVERY_CONTRACT]",
+    "This is an interactive UI delivery task.",
+    "A helper-only, data-only, validation-only, or serializable-view implementation does not satisfy the task.",
+    "Before editing: identify the existing UI framework and component conventions, the real render/integration point, and the user-interaction test seam.",
+    "Required delivery: a framework-native visible component; integration into the declared route or parent; user interaction and visible error rendering; component-level tests using the repository's UI test stack.",
+    "The completion summary must list the component path, integration path, interaction test path, and DOM assertions mapped to acceptance criteria. If any item is missing, do not claim completion.",
+    "[/INTERACTIVE_UI_DELIVERY_CONTRACT]",
+].join("\n");
+function isInteractiveUiTask(taskConfig) {
+    return taskConfig.capabilities?.includes("interactive-ui") ?? false;
+}
+function writerDeliveryContract(taskConfig) {
+    return isInteractiveUiTask(taskConfig)
+        ? INTERACTIVE_UI_DELIVERY_CONTRACT
+        : null;
+}
+function resolveWriterComplexity(taskConfig) {
+    if (isInteractiveUiTask(taskConfig))
+        return "HIGH";
+    if (taskConfig.complexity === "small")
+        return "MED";
+    if (taskConfig.complexity === "large")
+        return "HIGH";
+    return "MED";
+}
 const HYBRID_DEFAULTS = {
     executor: "pi",
     piBackend: "sdk-first",
@@ -437,11 +463,7 @@ export function buildStandardHybridDagFromTask(sources) {
     const implementPaths = resolveImplementPaths(taskConfig);
     const scoutPaths = deriveParallelScoutPaths(taskConfig);
     const scoutComplexity = mapTaskComplexity(taskConfig.complexity);
-    const implementComplexity = taskConfig.complexity === "small"
-        ? "MED"
-        : taskConfig.complexity === "large"
-            ? "HIGH"
-            : "MED";
+    const implementComplexity = resolveWriterComplexity(taskConfig);
     const writerExecutor = resolveImplementationExecutor(sources);
     const implementId = implementationNodeId(writerExecutor);
     const sourceContext = buildSourceContextBlock(sources);
@@ -543,8 +565,9 @@ export function buildStandardHybridDagFromTask(sources) {
                 subtask_prompt: [
                     "Implement the approved plan with minimal focused changes.",
                     "Stay within writeSet. Do not write root artifacts/** unless artifacts paths are explicitly declared in writeSet.",
+                    writerDeliveryContract(taskConfig),
                     sourceContext,
-                ].join("\n\n"),
+                ].filter((value) => Boolean(value)).join("\n\n"),
             },
             {
                 id: "verify-pi",
@@ -828,7 +851,11 @@ function buildRepairNode(sources) {
         role: "implementer",
         executor: writerExecutor,
         toolProfile: "write",
-        complexity: sources.taskConfig.complexity === "small" ? "MED" : "HIGH",
+        complexity: isInteractiveUiTask(sources.taskConfig)
+            ? "HIGH"
+            : sources.taskConfig.complexity === "small"
+                ? "MED"
+                : "HIGH",
         writePolicy: "exclusive",
         writeSet: implement.writeSet,
         allowedPaths: implement.allowedPaths,
@@ -838,8 +865,9 @@ function buildRepairNode(sources) {
             "If process-supervisor-pi returned VERDICT: request-revision, apply bounded fixes within writeSet addressing its REPAIR_ARTIFACT_JSON.fixScope and preserving REPAIR_ARTIFACT_JSON.invariant.",
             "Use REPAIR_ARTIFACT_JSON.failureClass and rootCause to choose the smallest repair strategy. Raw log is fallback evidence only when rawLogFallbackAllowed is true.",
             "If VERDICT: pass, return no-op with evidence. Re-run focused tests when you change code.",
+            writerDeliveryContract(sources.taskConfig),
             buildSourceContextBlock(sources),
-        ].join("\n\n"),
+        ].filter((value) => Boolean(value)).join("\n\n"),
     };
 }
 function buildHardVerifyNode(sources) {

package/docs/README.md

@@ -53,6 +53,7 @@
 - `templates/qa-report.md` — 验证与 QA 证据
 - `templates/worker-dogfood-setup.md` — 发布控制器下的真实 Worker sample setup 与 retry 纪律
 - `templates/worker-dogfood-evidence.md` — BE/FE/QA sample、Observe、morning report 与 coverage evidence 模板
+- `templates/interactive-ui-round2-experiment.md` — interactive UI prompt/model A/B/C 对照实验与统一指标模板
 - `templates/production-readiness-checklist.md` — 低/中风险单仓库 DAG readiness 检查清单
 - `templates/init-evolution-review.md` — 初始化能力演化审查报告模板
 - `templates/adr.md` — 架构决策记录（ADR）

package/docs/exec-plans/active/README.md

@@ -4,9 +4,9 @@
 
 源码仓库可在本 README 旁保留具体 active plan 文件。npm 包只携带本 README 作为目录契约，不复制 loop-agent 源码历史的 active plan；目标仓库自行生成 active plan。
 
-`2026-07-08-taskspec-worker-master.md` 的 7 个 step 已全部完成并归档到 `../completed/`。
+当前 active execution plan：
 
-当前没有 active execution plan 以外的进行中项。
+- `2026-07-10-release-0.6.0-nightly-drill.md` — 技术演练已完成；等待 0.6.0 发布 owner 对 Day-1 provisional remediation 作明确 ratify / revise / reject。
+- `2026-07-11-round-2-interactive-ui-productization.md` — 补齐 Round-1 owner gate，发布 interactive-ui controller，完成 React A/B/C、Round 2/3 扩量、CI/cron 和产品线模板/docs CI。
 
-- active：`2026-07-10-release-0.6.0-nightly-drill.md` — 发布 0.6.0，并用已发布控制器完成第 4 周首轮夜间 batch + 次日 morning report 决策闭环。
-- 已归档：`../completed/2026-07-10-next-stage-worker-evidence.md` 及更早的 TaskSpec/OBS 计划。
+- 已归档：`../completed/2026-07-11-interactive-ui-writer-routing.md`、`../completed/2026-07-10-next-stage-worker-evidence.md` 及更早计划。

package/docs/exec-plans/completed/README.md

@@ -8,6 +8,7 @@ npm 包携带本 README 作为目录契约。具体 completed plan 属于目标
 - [`2026-07-04-remove-level1-fallback.md`](2026-07-04-remove-level1-fallback.md) — 移除历史顺序 Level 1 fallback，runtime、文档与 command surface 收敛到 DAG 执行
 - [`2026-07-04-runtime-boundary-remediation.md`](2026-07-04-runtime-boundary-remediation.md) — 整合 CLI/skill/runtime 边界，抽出 DAG/Loop runtime seam，集中 harness store/guard 策略
 - [`2026-07-10-next-stage-worker-evidence.md`](2026-07-10-next-stage-worker-evidence.md) — Worker retry、EnvFailure、边界治理、初始化投影与真实 BE/FE/QA dogfood evidence 闭环
+- [`2026-07-11-interactive-ui-writer-routing.md`](2026-07-11-interactive-ui-writer-routing.md) — 新增 interactive-ui writer-only HIGH 路由、UI 交付契约、真实 React dogfood fixture 与 Round-2 A/B/C 实验入口
 - [`2026-07-04-dag-role-skill-alignment.md`](2026-07-04-dag-role-skill-alignment.md) — 对齐 DAG/Dynamic Workflow role 与 repo-local vetted skills，新增 strict skill audit
 - [`2026-07-06-production-readiness-hardening.md`](2026-07-06-production-readiness-hardening.md) — 冻结 Production Readiness v0.1，打磨 DAG 主路径 next steps、failure routing、doctor/report/failure handoff 与 dogfood 验证
 - [`2026-07-08-taskspec-schema-validate.md`](2026-07-08-taskspec-schema-validate.md) — 新增 TaskSpec v0.1 schema、三层校验器、risk→complexity 映射和 5 个 dogfood TaskSpec 样例

package/docs/init-surface.manifest.json

@@ -16,6 +16,7 @@
     "docs/templates/production-readiness-checklist.md",
     "docs/templates/worker-dogfood-setup.md",
     "docs/templates/worker-dogfood-evidence.md",
+    "docs/templates/interactive-ui-round2-experiment.md",
     "docs/templates/agent-dag.schema.json",
     "examples/example-dag.json",
     "skills/loop-agent/SKILL.md",
@@ -44,6 +45,7 @@
     "docs/templates/production-readiness-checklist.md",
     "docs/templates/worker-dogfood-setup.md",
     "docs/templates/worker-dogfood-evidence.md",
+    "docs/templates/interactive-ui-round2-experiment.md",
     "scripts/check-repo.sh",
     "scripts/ci-governance.sh",
     "scripts/ci-tests.sh",
@@ -88,6 +90,7 @@
     "docs/templates/production-readiness-checklist.md": "copied",
     "docs/templates/worker-dogfood-setup.md": "copied",
     "docs/templates/worker-dogfood-evidence.md": "copied",
+    "docs/templates/interactive-ui-round2-experiment.md": "copied",
     "scripts/check-repo.sh": "generated",
     "scripts/ci-governance.sh": "generated",
     "scripts/ci-tests.sh": "generated",

package/docs/templates/interactive-ui-round2-experiment.md

@@ -0,0 +1,66 @@
+# Interactive UI Round-2 A/B/C Experiment
+
+## Frozen inputs
+
+- Target repo / commit:
+- Controller version:
+- TaskSpec / acceptance hash:
+- Base DAG:
+- Model matrix:
+
+Prepare the fixture and three DAGs after installing a published controller that contains `interactive-ui` support:
+
+```bash
+bash scripts-local/setup-drill-round2-react.sh /tmp/drill-round2-react-target
+npm run build
+node scripts-local/prepare-round2-ui-experiment.mjs \
+  /tmp/drill-round2-react-target \
+  dogfood/features/F-2026-001/tasks/FE-001.yaml \
+  /tmp/round2-ui-experiment \
+  loop-agent
+node scripts-local/build-round2-ui-variants.mjs \
+  /tmp/round2-ui-experiment/round2-ui-base.json \
+  /tmp/round2-ui-experiment/variants
+```
+
+Validate every generated variant with the same frozen controller before running it. Use a unique run ID for A, B, and C; do not rewrite `executorModels`.
+
+## Variants
+
+| Variant | Writer prompt | Writer tier | Run ID | Result |
+| --- | --- | --- | --- | --- |
+| A | interactive-ui contract | MED | | |
+| B | default contract | HIGH | | |
+| C | interactive-ui contract | HIGH | | |
+
+## Metrics
+
+| Metric | A | B | C |
+| --- | --- | --- | --- |
+| First-pass review gate pass | | | |
+| Framework-native `.tsx` component | | | |
+| Route/parent integration | | | |
+| DOM interaction tests | | | |
+| Helper-only escape | | | |
+| Duration | | | |
+| Tokens | | | |
+
+## Required evidence per run
+
+- DAG JSON and run ID
+- implement/repair writer model and prompt profile
+- changed component path
+- integration path
+- interaction test path
+- DOM assertions mapped to AC-FE-001
+- review verdict and deterministic test output
+- diff boundary audit
+
+## Decision
+
+- Production default:
+- Evidence:
+- Cost/quality trade-off:
+- Follow-up:
+
+Do not conclude from a single run when provider or environment failures occurred. Re-run the affected variant with the same frozen inputs and a new run ID.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tea-agent/loop-agent",
-  "version": "0.6.0",
+  "version": "0.7.0",
   "type": "module",
   "bin": {
     "loop-agent": "bin/loop-agent.js",

package/skills/loop-agent/references/command-reference.md

@@ -377,7 +377,7 @@ agent-worker report morning --repo <repo-root> [--batch-run-id <id>] [--output <
 
 语义要点：
 
-- TaskSpec 声明单个任务的业务上下文、`risk_level`、验收与 verify 边界；`risk_level` 被确定性映射到 task complexity。AcceptanceSpec / TaskGraphSpec 声明跨任务验收引用与依赖图，ready queue 决定可运行任务并检测未知依赖/环/文件一致性。
+- TaskSpec 声明单个任务的业务上下文、`risk_level`、可选 `capabilities`、验收与 verify 边界；`risk_level` 被确定性映射到 task complexity。`capabilities: [interactive-ui]` 不改变风险或治理 profile，只把 implement/repair writer 路由到 HIGH，并注入禁止 helper-only 逃逸的真实 UI 交付契约。AcceptanceSpec / TaskGraphSpec 声明跨任务验收引用与依赖图，ready queue 决定可运行任务并检测未知依赖/环/文件一致性。
 - 业务 type（`backend-feature`/`frontend-feature`/`qa-testcode` 等）是产品线 profile，不能直接传给 `loop-agent dag run-task --profile`；Worker 会映射到 `auto`/`minimal`/`standard`/`reviewed`/`supervised`。
 - materializer 把 TaskSpec 物化为 `.harness/tasks/<task-id>`：`source_docs` 原样进入 `source/references/`，派生 `需求.md` 带权威声明、Source Docs/hash 追溯，以及 `acceptance_refs` 短摘要；随后 Worker 调用 `dag run-task` / `dag validate` / `run-dag` / `dag report`。review 节点须对照 references + 派生契约 + 实现。成功路径走 `promote-run` + `closeout task`；失败路径收集 `dag doctor` / `dag closeout-draft` evidence，写入目标 repo 的 `.task-pool/failure-handoffs/`（不写 `.harness/`，因为 `dag closeout-draft` 会拒绝在 completed facts 之外写入）。
 - Worker runtime state 落在目标 repo 的 `.task-pool/`（artifacts、JSONL/state、晨报、failure handoffs），与 `.harness/` 分离；`.task-pool/` 默认被 `.gitignore` 忽略。