@agile-team/wl-skills-ui 1.14.0 → 1.16.0

package/CHANGELOG.md

@@ -6,6 +6,30 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [1.16.0] - 2026-10-08
+
+### Added
+
+- 新增公开集成协议 describe/request 统一信封：能力清单与路由语料经真实打包产物验证，供外部宿主与适配器按公开契约接入。
+
+### Fixed
+
+- 输入校验补齐：context 为 null、targets 为空等边界输入返回可解释拒绝，不再产生歧义结果。
+- MCP 目录对齐真实 `tools/list` 的 18 项，修复目录声明与实际暴露不一致。
+
+## [1.15.0] - 2026-10-07
+
+### Added
+
+- 新增 task/route/explain/status/doctor-host CLI 与 5 个 MCP 工具，任务计划与真实执行回执分开；支持跨已安装适用包复用 runId，无兄弟包运行依赖。
+- 新增本包独有 Codex `.agents/skills/wl-skills-ui/SKILL.md` 薄入口，按需读取 canonical Skill；缺资产或未发布显式技能报告 gap，静态入口诊断不冒充宿主实际发现。
+- scan/check/fix/all 与 MCP、Vite 依赖检查记录执行/验证状态、实际检查文件、规则回调次数、规则/配置快照与过期状态。未执行、零文件、跳过或人工复核不会假绿。
+
+### Fixed
+
+- 实际 Vue SFC 解析错误形成不可被 only/skip 隐藏的 SFC_PARSE 诊断；fast/fallback 不声称语法验证，脚本类型编译仍需项目独立 typecheck。
+- 不适用/缺上下文任务不附加全量 UI 约束；Markdown 无问题文件占比不再称为规范覆盖率，零文件明确未检查。Vite 插件成功也明确仅验证依赖组合、源码规则未运行。
+
 ## [1.14.0] - 2026-10-07
 
 ### Added

package/README.md

@@ -1,6 +1,6 @@
 # @agile-team/wl-skills-ui
 
-**企业级 UI 一致性与 AI 精确治理工具包 v1.14.0** — 面向 Vue 3 + Element Plus 的设计令牌、能力 Profile、样式皮肤、页面骨架、运行时能力、静态扫描、确定性修复、AI Skill、MCP 与脱敏 UI 模板契约。
+**企业级 UI 一致性与 AI 精确治理工具包 v1.16.0** — 面向 Vue 3 + Element Plus 的设计令牌、能力 Profile、样式皮肤、页面骨架、运行时能力、静态扫描、确定性修复、AI Skill、MCP 与脱敏 UI 模板契约。
 
 它把“视觉统一、问题识别、机械修复、项目适配和样例沉淀”拆成可验证、可组合的工程能力。脚本、规则和 MCP 先完成确定性工作，AI 只处理业务语义与人工决策，从而减少重复读源码、上下文 token 和误修改。
 
@@ -15,13 +15,16 @@
 | 低成本治理老项目 | Skin 模式、Base*/jh-*/C_*/AG Grid 兼容层、定制页豁免 | 不改封装源码也能统一大部分视觉，避免一次性重构风险 |
 | 阻止 UI 回归 | 40 条 R-rule、接入检查、漂移基线、Git 增量扫描、CI 门禁 | 表头/内容不共轴、按钮无图标、日期弹层全屏、密集表单失衡等问题在提交前暴露 |
 | 安全机械修复 | Profile、only/skip、dry-run 计划哈希、快照、复检与回滚 | 自动化只处理可证明安全且已批准的动作，不让 AI 猜测批改 |
-| 降低 AI token | summary、compact-v2 分页、changed-only、Skill 路由、13 个 MCP Tool | AI 先读总览和结构化事实，只在必要时获取有界明细或局部源码 |
+| 降低 AI token | summary、compact-v2 分页、changed-only、Skill 路由、18 个 MCP Tool | AI 先读总览和结构化事实，只在必要时获取有界明细或局部源码 |
 | 沉淀领域样例 | `ui-contract` extract / validate / match | 页面转为脱敏语义 JSON，不复制整页代码、接口、字段值或原始文案 |
 | 兼容多代项目 | EP 2.2+jh-ui 与原生 EP 2.7+ DOM 契约、Vite/doctor 检查 | 支持声明有 fixture 和真实浏览器证据，不靠文档口头约定 |
 | 降低运行时开销 | Shared Observer Hub、可选 guard、幂等安装与完整卸载 | 全部 guard 共用 1 个 MutationObserver + 1 个 ResizeObserver |
 
 适合多项目 Vue 3 企业后台、存量系统视觉治理、AI 辅助开发和领域页面模式沉淀。运行时最低要求：Node.js 22、Vue 3.2、Element Plus 2.2。
 
+
+每次任务先运行 `wl-ui task "任务" --target <path>`，按选定技能执行工具，最后用 `wl-ui status --run-id <id>` 查看实际执行与验证状态；安装或匹配不会自动变成检查通过。详见[任务判定与真实检查回执](docs/task-observability.md)。
+
 ## 快速开始
 
 ### 1. 安装
@@ -91,6 +94,23 @@ CLI、安装器和 MCP 无需安装其他 wl-skills 包或统一客户端即可
 
 `init` 新插入的 HTML tokens link 使用独立边界标记并登记贡献，`clean` 只移除未修改的本包链接，保留项目其他 HTML。历史未标记的链接缺少归属证据，升级时仍作为项目配置保留；SCSS 与 runtime 接入只给出建议，不自动修改或认领项目已有 import。
 
+## 公开集成协议 v1
+
+供 harness、客户端或其他编排层以统一 JSON 入口调用本包判定与状态能力（业务执行仍走本包 CLI/MCP，不因协议新增兄弟包依赖）。
+
+```bash
+wl-ui protocol describe --json                      # 能力目录、五操作映射、请求/响应 Schema、错误码
+wl-ui protocol request --input-file request.json --json
+```
+
+- 操作：`route` / `explain`（只读判定）、`task`（判定并持久化计划，尚未执行）、`status`（回查本包执行回执）、`doctor-host`（宿主入口静态诊断）。
+- 请求字段：`protocolVersion`(1)、`operation`、`requestId`、`runId`、`projectRoot`、`task`、`targets`、`skill`、`host`+ `profile`。
+- 统一信封：`{ ok, result | error, diagnostics }`；机器结果与诊断分离；非法输入在触达执行器前判 `invalid-input`，不写入任何记录。
+- 错误码：`unsupported-protocol` / `unknown-operation` / `missing-input` / `invalid-input` / `internal-error`；请求失败（exit 2）时 stdout 仍为可解析 JSON 信封。
+- **`ok=true` 只表示协议调用成功，不代表业务验证通过**；业务状态以 `result` 内 `validationStatus` / `executionStatus` 为准（仅计划时为 `unverified` / `not-executed`）。
+- 副作用：仅 `task` 写入本包自己的 runs 目录（见 describe 中各操作 `sideEffects`）；其余操作只读。
+- 协议实现为本包内置的同源快照 `integration-protocol.cjs`，五包一致；跨包一致性由 conformance `integration-protocol.check.cjs` 验收。
+
 ## 能力地图
 
 | 层级 | 能力 | 典型内容 | 是否需要业务源码 |
@@ -480,7 +500,7 @@ npx wl-ui doctor --project . --print-overrides
 
 ## 版本与进一步阅读
 
-当前 v1.14.0：明确独立能力边界，安装、更新、清理按本包贡献保护用户及其他包内容；共享文件采用单文件原子替换，支持历史路由与 `.clinerules` 目录迁移。v1.13.2：R044 审查多列表单超宽标签，R040 识别自定义 `stepper`，均供人工复核。v1.13.0：新增 `native-jh-ag` Profile、显式 Profile 兼容承诺与联邦 AG 依赖识别。
+当前 v1.16.0：公开集成协议 describe/request 统一信封（能力清单与路由语料经真实打包产物验证）、输入校验收口与 MCP 目录对齐真实 tools/list。v1.15.0：任务判定、真实执行/验证回执、实际检查范围和过期检测；Codex 原生薄入口按需加载 canonical Skill。确定性路由覆盖已发布目录与显式缺口，不保证识别所有自然语言。v1.14.0：明确独立能力边界，安装、更新、清理按本包贡献保护用户及其他包内容；共享文件采用单文件原子替换，支持历史路由与 `.clinerules` 目录迁移。v1.13.2：R044 审查多列表单超宽标签，R040 识别自定义 `stepper`，均供人工复核。v1.13.0：新增 `native-jh-ag` Profile、显式 Profile 兼容承诺与联邦 AG 依赖识别。
 
 - [版本摘要](docs/version-history.md)
 - [完整 Changelog](CHANGELOG.md)

package/SKILL.md

@@ -14,6 +14,8 @@ do not reconstruct them from memory or duplicate their wording.
 
 ## Start here
 
+Start every task with `npx wl-ui task "<task>" --target <path>`. Read the selected canonical Skill and required rules only after resolving ambiguous/gap/needs-context or missing own assets. A match is a plan, not evidence of execution. Use the returned `--run-id` (or `WL_TASK_RUN_ID`) for real CLI/MCP tools and finish with `npx wl-ui status --run-id <id>`. Reuse the same ID across installed applicable packages without installing unused sibling packages. Report execution and validation separately, including skipped checks and stale evidence; host discovery/reading remain unverified without host events.
+
 ```bash
 # Small summary first; does not modify files
 npx wl-ui scan --project . --target src --output summary --parser auto

package/bin/__tests__/protocol-cli.test.mjs

@@ -0,0 +1,122 @@
+/**
+ * protocol-cli.test.mjs — ui 公开集成协议统一信封与接线测试
+ */
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync } from "node:fs";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { spawnSync } from "node:child_process";
+import { fileURLToPath } from "node:url";
+import { createRequire } from "node:module";
+
+const require = createRequire(import.meta.url);
+const here = path.dirname(fileURLToPath(import.meta.url));
+const { protocol, runOperation } = await import("../protocol-cli.mjs");
+const pkg = require("../../package.json");
+const BIN = path.join(here, "..", "wl-ui.js");
+
+function tempRoot() {
+  return mkdtempSync(path.join(tmpdir(), "wl-ui-protocol-"));
+}
+
+test("describe 返回协议版本、能力目录与五个统一操作", () => {
+  const described = protocol.describe();
+  assert.equal(described.protocolVersion, 1);
+  assert.equal(described.package, pkg.name);
+  assert.ok(Array.isArray(described.capabilities) && described.capabilities.length > 0);
+  assert.deepEqual(described.operations.map((operation) => operation.id), ["route", "explain", "task", "status", "doctor-host"]);
+  assert.ok(described.errorCodes.includes("unsupported-protocol"));
+});
+
+test("route 只读判定返回六类状态之一", () => {
+  const envelope = protocol.request({ operation: "route", projectRoot: tempRoot(), task: "统一这个页面的表格视觉" }, runOperation);
+  assert.equal(envelope.ok, true);
+  const status = envelope.result.decision ? envelope.result.decision.status : envelope.result.status;
+  assert.ok(["matched", "baseline", "ambiguous", "gap", "not-applicable", "needs-context"].includes(status));
+});
+
+test("task 持久化返回 runId 且 status 可回查", () => {
+  const root = tempRoot();
+  const planned = protocol.request({ operation: "task", projectRoot: root, task: "扫描 src/views 视觉问题" }, runOperation);
+  assert.equal(planned.ok, true);
+  assert.ok(planned.result.runId);
+  const status = protocol.request({ operation: "status", projectRoot: root, runId: planned.result.runId }, runOperation);
+  assert.equal(status.ok, true);
+  assert.equal(status.result.runId, planned.result.runId);
+  assert.equal(status.result.executionStatus, "not-executed");
+});
+
+test("doctor-host 按指定 host 诊断", () => {
+  const envelope = protocol.request({ operation: "doctor-host", projectRoot: tempRoot(), host: "claude" }, runOperation);
+  assert.equal(envelope.ok, true);
+  assert.equal(envelope.result.host, "claude");
+});
+
+test("协议错误码：missing-input / unknown-operation / invalid-input / unsupported-protocol", () => {
+  assert.equal(protocol.request({ operation: "route" }, runOperation).error.code, "missing-input");
+  assert.equal(protocol.request({ operation: "skin" }, runOperation).error.code, "unknown-operation");
+  assert.equal(protocol.request(42, runOperation).error.code, "invalid-input");
+  assert.equal(protocol.request({ protocolVersion: 3, operation: "route", task: "x" }, runOperation).error.code, "unsupported-protocol");
+});
+
+test("边界输入校验：非法类型在触达执行器前判 invalid-input（独立复验缺陷回归）", () => {
+  const invalidPayloads = [
+    { operation: "task", task: true },
+    { operation: "task", task: { text: "bad" } },
+    { operation: "task", task: "检查目标", targets: [null, 42, {}] },
+    { operation: "route", task: "检查目标", projectRoot: 42 },
+    { operation: "task", task: "检查目标", runId: {} },
+  ];
+  for (const payload of invalidPayloads) {
+    const envelope = protocol.request(payload, () => { throw new Error("不应触达执行器"); });
+    assert.equal(envelope.ok, false);
+    assert.equal(envelope.error.code, "invalid-input");
+    assert.ok(envelope.error.field);
+  }
+});
+
+test("CLI 缺 --input-file 时 stdout 输出 missing-input JSON 信封", () => {
+  const run = spawnSync(process.execPath, [BIN, "protocol", "request"], { encoding: "utf8" });
+  assert.equal(run.status, 2);
+  const envelope = JSON.parse(run.stdout);
+  assert.equal(envelope.ok, false);
+  assert.equal(envelope.error.code, "missing-input");
+  assert.equal(envelope.error.field, "input-file");
+});
+
+test("CLI protocol describe / request 全链路", () => {
+  const describe = spawnSync(process.execPath, [BIN, "protocol", "describe"], { encoding: "utf8" });
+  assert.equal(describe.status, 0);
+  assert.equal(JSON.parse(describe.stdout).package, "@agile-team/wl-skills-ui");
+  const root = tempRoot();
+  const file = path.join(root, "req.json");
+  const { writeFileSync } = require("node:fs");
+  writeFileSync(file, JSON.stringify({ operation: "doctor-host", projectRoot: root, host: "copilot" }));
+  const request = spawnSync(process.execPath, [BIN, "protocol", "request", "--input-file", file], { encoding: "utf8" });
+  assert.equal(request.status, 0);
+  assert.equal(JSON.parse(request.stdout).result.host, "copilot");
+});
+
+
+test("context null 判 invalid-input 且零写入（CLI 回归）", () => {
+  const root = tempRoot();
+  const file = path.join(root, "req.json");
+  const { writeFileSync, readdirSync } = require("node:fs");
+  writeFileSync(file, JSON.stringify({ operation: "task", projectRoot: root, task: "扫描 src/views 视觉问题", runId: "ctx-null", context: null }));
+  const run = spawnSync(process.execPath, [BIN, "protocol", "request", "--input-file", file], { encoding: "utf8" });
+  assert.equal(run.status, 2);
+  const envelope = JSON.parse(run.stdout);
+  assert.equal(envelope.ok, false);
+  assert.equal(envelope.error.code, "invalid-input");
+  assert.equal(envelope.error.field, "context");
+  assert.equal(readdirSync(root).filter((name) => name.startsWith(".")).length, 0, "不得写入任何任务记录");
+});
+
+test("targets 空数组与 Schema 一致（运行时接受空范围）", () => {
+  const root = tempRoot();
+  const envelope = protocol.request({ operation: "route", projectRoot: root, task: "扫描 src/views 视觉问题", targets: [] }, runOperation);
+  assert.equal(envelope.ok, true);
+  const described = protocol.describe();
+  assert.equal(described.schemas.request.properties.targets.minItems, undefined, "Schema 不得要求非空 targets");
+});

package/bin/__tests__/protocol-evidence.test.mjs

@@ -0,0 +1,50 @@
+/**
+ * protocol-evidence.test.mjs — ui 证据闭环严格断言（不得兜底通过）
+ */
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync } from "node:fs";
+import { tmpdir } from "node:os";
+import path from "node:path";
+
+const { protocol, runOperation } = await import("../protocol-cli.mjs");
+
+function tempRoot() {
+  return mkdtempSync(path.join(tmpdir(), "wl-ui-evidence-"));
+}
+
+test("status 严格回查同一 runId：字段精确、不兜底", () => {
+  const root = tempRoot();
+  const planned = protocol.request({ operation: "task", projectRoot: root, task: "统一表格视觉", runId: "evidence-run-a" }, runOperation);
+  assert.equal(planned.ok, true);
+  assert.equal(planned.result.runId, "evidence-run-a");
+  assert.equal(planned.result.executionStatus, "not-executed");
+  assert.equal(planned.result.validationStatus, "unverified");
+  const status = protocol.request({ operation: "status", projectRoot: root, runId: "evidence-run-a" }, runOperation);
+  assert.equal(status.result.runId, "evidence-run-a");
+  assert.equal(status.result.executionStatus, "not-executed");
+});
+
+test("混用 runId 不得串记录", () => {
+  const root = tempRoot();
+  protocol.request({ operation: "task", projectRoot: root, task: "任务A统一表格视觉", runId: "evidence-run-a" }, runOperation);
+  protocol.request({ operation: "task", projectRoot: root, task: "任务B统一表格视觉", runId: "evidence-run-b" }, runOperation);
+  const statusB = protocol.request({ operation: "status", projectRoot: root, runId: "evidence-run-b" }, runOperation);
+  assert.equal(statusB.result.runId, "evidence-run-b");
+  assert.equal(JSON.stringify(statusB.result).includes("任务A"), false);
+});
+
+test("不存在的 runId 不得伪造成功记录", () => {
+  const root = tempRoot();
+  const status = protocol.request({ operation: "status", projectRoot: root, runId: "no-such-run" }, runOperation);
+  const serialized = JSON.stringify(status.result);
+  assert.equal(serialized.includes('"validationStatus":"passed"'), false);
+  assert.equal(serialized.includes('"executionStatus":"completed"'), false);
+});
+
+test("空检查集不得判定为通过", () => {
+  const root = tempRoot();
+  const planned = protocol.request({ operation: "task", projectRoot: root, task: "统一表格视觉", runId: "empty-run" }, runOperation);
+  assert.notEqual(planned.result.validationStatus, "passed");
+  assert.equal(planned.result.executionStatus, "not-executed");
+});

package/bin/__tests__/protocol-routing-cases.json

@@ -0,0 +1,105 @@
+[
+  {
+    "task": "统一表格视觉",
+    "skill": "element/el-table"
+  },
+  {
+    "task": "查询表单密集审计",
+    "skill": "element/el-form"
+  },
+  {
+    "task": "弹窗样式对齐",
+    "skill": "element/el-dialog"
+  },
+  {
+    "task": "状态标签颜色统一",
+    "skill": "element/el-tag"
+  },
+  {
+    "task": "列表布局骨架",
+    "skill": "layouts/list-page"
+  },
+  {
+    "task": "AG Grid 适配",
+    "skill": "vendors/ag-grid"
+  },
+  {
+    "task": "UI扫描",
+    "skill": "ops/scan"
+  },
+  {
+    "task": "UI修复统一视觉",
+    "skill": "ops/fix"
+  },
+  {
+    "task": "UI迁移迁移tokens",
+    "skill": "ops/migrate"
+  },
+  {
+    "task": "设计令牌接入消除颜色硬编码",
+    "skill": "runtime/design-tokens"
+  },
+  {
+    "task": "写诗",
+    "skill": null,
+    "status": "not-applicable"
+  },
+  {
+    "task": "生成测试用例",
+    "skill": null,
+    "status": "needs-context"
+  },
+  {
+    "task": "检查",
+    "skill": null,
+    "status": "needs-context"
+  },
+  {
+    "task": "整理页签和卡片组件",
+    "skill": "element/component-family"
+  },
+  {
+    "task": "详情布局整改",
+    "skill": "layouts/detail-page"
+  },
+  {
+    "task": "调整表单弹窗布局",
+    "skill": "layouts/form-dialog"
+  },
+  {
+    "task": "树列表布局",
+    "skill": "layouts/tree-list"
+  },
+  {
+    "task": "做一次UI审计",
+    "skill": "ops/audit"
+  },
+  {
+    "task": "runtime迁移",
+    "skill": "runtime/migration"
+  },
+  {
+    "task": "样式统一风格对齐",
+    "skill": "runtime/style-align"
+  },
+  {
+    "task": "BaseTable 适配",
+    "skill": "vendors/base-table"
+  },
+  {
+    "task": "C组件用法审计",
+    "skill": "vendors/c-components"
+  },
+  {
+    "task": "自定义封装组件治理",
+    "skill": "vendors/custom-wrappers"
+  },
+  {
+    "task": "jh组件接入",
+    "skill": "vendors/jh-components"
+  },
+  {
+    "task": "未知封装组件排查",
+    "skill": "vendors/unknown-wrapper"
+  }
+]

package/bin/__tests__/protocol-routing.test.mjs

@@ -0,0 +1,63 @@
+/**
+ * protocol-routing.test.mjs — ui 路由准确性逐技能语料（预期经人工复核冻结）
+ * ui 是依赖型包：安装态通过真实打包产物 npm 安装到隔离项目验证。
+ */
+import { test, before } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { spawnSync } from "node:child_process";
+import { fileURLToPath } from "node:url";
+
+const here = path.dirname(fileURLToPath(import.meta.url));
+const { protocol, runOperation } = await import("../protocol-cli.mjs");
+const cases = JSON.parse((await import("node:fs")).readFileSync(path.join(here, "protocol-routing-cases.json"), "utf8"));
+const PKG_ROOT = path.join(here, "..", "..");
+
+const state = { installedRoot: null, tarball: null };
+
+function tempRoot() {
+  return mkdtempSync(path.join(tmpdir(), "wl-ui-routing-"));
+}
+
+function routeAt(projectRoot, task) {
+  return protocol.request({ operation: "route", projectRoot, task }, runOperation);
+}
+
+before(() => {
+  const packDir = tempRoot();
+  const pack = spawnSync("npm", ["pack", "--pack-destination", packDir, "--ignore-scripts"], { cwd: PKG_ROOT, encoding: "utf8", timeout: 180000 });
+  assert.equal(pack.status, 0, pack.stderr);
+  state.tarball = path.join(packDir, pack.stdout.trim().split("\n").pop());
+  state.installedRoot = tempRoot();
+  assert.equal(spawnSync("npm", ["init", "-y"], { cwd: state.installedRoot, encoding: "utf8" }).status, 0);
+  const install = spawnSync("npm", ["i", state.tarball], { cwd: state.installedRoot, encoding: "utf8", timeout: 180000 });
+  assert.equal(install.status, 0, install.stderr);
+}, 600000);
+
+for (const item of cases) {
+  const expectedBare = item.skill ? "gap" : item.status;
+  const expectedInstalled = item.installedStatus || (item.skill ? "matched" : item.status);
+
+  test(`未安装：「${item.task}」→ ${expectedBare}${item.skill ? ` + ${item.skill}` : ""}`, () => {
+    const envelope = routeAt(tempRoot(), item.task);
+    assert.equal(envelope.ok, true);
+    const decision = envelope.result.decision || envelope.result;
+    assert.equal(decision.status, expectedBare);
+    if (item.skill) assert.ok(decision.selectedSkills.includes(item.skill), `应选中 ${item.skill}，实际 ${decision.selectedSkills}`);
+    else assert.equal((decision.selectedSkills || []).length, 0);
+  });
+
+  test(`已安装（真实打包产物）：「${item.task}」→ ${expectedInstalled}${item.skill ? ` + ${item.skill}` : ""}`, () => {
+    const envelope = routeAt(state.installedRoot, item.task);
+    assert.equal(envelope.ok, true);
+    const decision = envelope.result.decision || envelope.result;
+    assert.equal(decision.status, expectedInstalled);
+    if (item.skill) assert.ok(decision.selectedSkills.includes(item.skill));
+  });
+}
+
+test.after(() => {
+  if (state.tarball) rmSync(path.dirname(state.tarball), { recursive: true, force: true });
+});

package/bin/capabilities.json

@@ -20,6 +20,27 @@
         "runtime-renderers",
         "visual-findings"
       ]
+    },
+    {
+      "id": "ui.task-observability",
+      "intents": [
+        "route",
+        "explain",
+        "record-execution",
+        "status",
+        "diagnose-host"
+      ],
+      "inputs": [
+        "task-description",
+        "target-paths",
+        "actual-tool-results"
+      ],
+      "outputs": [
+        "task-decision",
+        "execution-receipts",
+        "gap-proposals",
+        "host-entry-diagnostics"
+      ]
     }
   ],
   "standalone": {
@@ -38,6 +59,15 @@
     "optionalTraceInput": "design-model",
     "unknownProtocol": "reject-or-explicit-migration",
     "apiAuthority": "validated-effective-api-manifest",
-    "sharedWrites": "own-block-or-json-entry-only"
+    "sharedWrites": "own-block-or-json-entry-only",
+    "observability": {
+      "recordVersion": 1,
+      "schema": "bin/task-observability.schema.json",
+      "storage": ".wl-skills-ui/runs",
+      "runCorrelation": "explicit-runId-or-WL_TASK_RUN_ID",
+      "eventPublication": "immutable-per-package-events",
+      "hostDiscovery": "unverified-without-host-events",
+      "gapPolicy": "propose-review-before-changing-business-rules"
+    }
   }
 }