compforge · qiankunli · Aug 19, 2026 · Aug 18, 2026
diff --git a/.agents/plugins/marketplace.json b/.agents/plugins/marketplace.json
@@ -3,7 +3,7 @@
   "owner": {
     "name": "compforge"
   },
-  "description": "Cross-CLI developer-workflow marketplace for devloop and code-taste; example remains a layout placeholder.",
+  "description": "Cross-CLI developer-workflow marketplace for devloop, code-taste, and quality; example remains a layout placeholder.",
   "plugins": [
     {
       "name": "devloop",
@@ -41,6 +41,24 @@
       "category": "Developer Workflow",
       "keywords": ["architecture", "code-review", "design", "maintainability", "naming", "refactoring"]
     },
+    {
+      "name": "quality",
+      "source": {
+        "source": "local",
+        "path": "./quality"
+      },
+      "description": "Evidence-driven quality skills that discover and run project-owned verification capabilities.",
+      "author": {
+        "name": "compforge",
+        "url": "https://github.com/compforge"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Developer Workflow",
+      "keywords": ["quality", "e2e", "testing", "verification", "evidence"]
+    },
     {
       "name": "example",
       "source": {

diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
@@ -3,7 +3,7 @@
   "owner": {
     "name": "compforge"
   },
-  "description": "Cross-CLI developer-workflow marketplace for devloop and code-taste; example remains a layout placeholder.",
+  "description": "Cross-CLI developer-workflow marketplace for devloop, code-taste, and quality; example remains a layout placeholder.",
   "plugins": [
     {
       "name": "devloop",
@@ -23,6 +23,15 @@
         "url": "https://github.com/compforge"
       }
     },
+    {
+      "name": "quality",
+      "source": "./quality",
+      "description": "Evidence-driven quality skills that discover and run project-owned verification capabilities.",
+      "author": {
+        "name": "compforge",
+        "url": "https://github.com/compforge"
+      }
+    },
     {
       "name": "example",
       "source": "./example",

diff --git a/AGENTS.md b/AGENTS.md
@@ -1,6 +1,8 @@
 # devloop
 
-跨 CLI 的 plugin marketplace。一个 marketplace 仓库 + 多个 plugin 子目录 + 各 CLI 各自的 manifest。本仓库**不是单 plugin 仓库**——根目录的 `devloop/` / `code-taste/` / `example/` 等子目录每个都是独立 plugin。`devloop` 承载开发闭环，`code-taste` 提供设计判断 skill，`example` 是占位示例。
+devloop 是跨 CLI 的 plugin marketplace，根目录下每个 plugin 独立交付和演进。`devloop/devloop`
+是首个 Plugin，承载开发闭环；`code-taste` 提供设计判断 skill，`quality` 提供基于执行证据的质量
+评估 skill，`example` 是占位示例。
 
 ---
 
@@ -51,6 +53,12 @@ devloop/                              # ← 仓库根（marketplace）
 │   ├── skills/code-taste/            #     共享 skill 与分层 references
 │   └── README.md
 │
+├── quality/                          # plugin: 基于项目真实测试资产的质量评估（skill-only）
+│   ├── .claude-plugin/plugin.json    #     Claude manifest
+│   ├── .codex-plugin/plugin.json     #     Codex manifest
+│   ├── skills/e2e/                   #     发现、执行并解释项目已有 E2E 能力
+│   └── README.md
+│
 ├── example/                          # plugin: 占位演示，证明这是多 plugin marketplace
 │   ├── .claude-plugin/plugin.json    #     Claude ✅
 │   ├── .codex-plugin/plugin.json     #     Codex  ✅

diff --git a/README.md b/README.md
@@ -7,7 +7,8 @@
 This repository is a cross-CLI plugin marketplace. Its flagship plugin, `devloop`, helps Claude Code and Codex complete branch-based development without losing control of repository state, validation, Git mutations, or concurrent sessions.
 
 The marketplace also distributes `code-taste`, a skill-only plugin for architecture, boundaries,
-naming, maintainability, documentation, and code review.
+naming, maintainability, documentation, and code review, plus `quality`, a skill-only plugin for
+evidence-driven project quality evaluation.
 
 ```text
 enter repo → start branch → develop → normalize → lint ∥ test → commit → push → PR/MR → human merge
@@ -119,6 +120,7 @@ Start a new session after updating so the runtime reloads hooks and skills. User
 |---|---|---|
 | `devloop` | Controlled PR/MR development lifecycle for Claude Code and Codex | [README](./devloop/README.md) |
 | `code-taste` | Engineering judgment for architecture, boundaries, naming, maintainability, and review | [README](./code-taste/README.md) |
+| `quality` | Discover and run project-owned quality capabilities; initially E2E | [README](./quality/README.md) |
 | `example` | Placeholder demonstrating the multi-plugin marketplace layout | [README](./example/README.md) |
 
 ## Documentation

diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -2,7 +2,7 @@
 
 # devloop
 
-**为 AI 编码管理一条受控的 PR/MR 开发生命周期。** 本仓库是跨 CLI 的 plugin marketplace；主力 plugin `devloop` 引导 Claude Code 和 Codex 完成进入 repo、基于 branch 开发、按受影响 Component 验证、commit/push 与创建 PR/MR，最终 merge 始终留给人。**GitHub（PR）与 GitLab（MR）都支持**，按 repo 的 origin 自动识别；`code-taste` 作为独立 skill-only plugin 提供架构、边界、命名与可维护性判断，`example` 保留为占位示例。
+**为 AI 编码管理一条受控的 PR/MR 开发生命周期。** 本仓库是跨 CLI 的 plugin marketplace；主力 plugin `devloop` 引导 Claude Code 和 Codex 完成进入 repo、基于 branch 开发、按受影响 Component 验证、commit/push 与创建 PR/MR，最终 merge 始终留给人。**GitHub（PR）与 GitLab（MR）都支持**，按 repo 的 origin 自动识别；`code-taste` 作为独立 skill-only plugin 提供架构、边界、命名与可维护性判断，`quality` 提供基于执行证据的项目质量评估 skill，`example` 保留为占位示例。
 
 > 当前支持 Claude Code 和 Codex。设计理念 / 架构见 [AGENTS.md](./AGENTS.md)，各 plugin 细节见各自目录的 README。
 
@@ -133,6 +133,7 @@ opencode 仍是占位，等其 plugin / hook 协议接入后再启用。
 |--------|------|--------|
 | `devloop` | 面向 AI 编码的受控 PR/MR 生命周期：repo/branch 入口、受影响 Component 验证、commit/push/PR、实时状态与执行级护栏（Claude + Codex；GitHub + GitLab） | [devloop/README.md](./devloop/README.md) |
 | `code-taste` | 架构、边界、命名、可维护性、文档与 Code Review 判断 | [code-taste/README.md](./code-taste/README.md) |
+| `quality` | 发现并执行项目已有质量能力，首个 skill 为 E2E | [quality/README.md](./quality/README.md) |
 | `example` | 占位 plugin，演示多 plugin marketplace 结构 | [example/README.md](./example/README.md) |
 
 ## 新增 plugin

diff --git a/quality/.claude-plugin/plugin.json b/quality/.claude-plugin/plugin.json
@@ -0,0 +1,9 @@
+{
+  "name": "quality",
+  "version": "0.0.1",
+  "description": "Evidence-driven quality skills that discover and run project-owned verification capabilities.",
+  "author": {
+    "name": "compforge",
+    "url": "https://github.com/compforge"
+  }
+}
diff --git a/quality/.codex-plugin/plugin.json b/quality/.codex-plugin/plugin.json
@@ -0,0 +1,30 @@
+{
+  "name": "quality",
+  "version": "0.0.1",
+  "description": "Evidence-driven quality skills that discover and run project-owned verification capabilities.",
+  "author": {
+    "name": "compforge",
+    "url": "https://github.com/compforge"
+  },
+  "license": "MIT",
+  "keywords": [
+    "quality",
+    "e2e",
+    "testing",
+    "verification",
+    "evidence"
+  ],
+  "skills": "./skills/",
+  "interface": {
+    "displayName": "Quality",
+    "shortDescription": "Evidence-driven project quality evaluation",
+    "longDescription": "Helps coding agents discover, run, and interpret project-owned quality capabilities without inventing missing tests or treating unsupported areas as passed.",
+    "defaultPrompt": "Use the E2E skill to discover, run, and report this project's existing end-to-end tests.",
+    "developerName": "compforge",
+    "category": "Developer Workflow",
+    "capabilities": [
+      "Read",
+      "Interactive"
+    ]
+  }
+}
diff --git a/quality/README.md b/quality/README.md
@@ -0,0 +1,49 @@
+# Quality
+
+Quality is a skill-only plugin that helps coding agents evaluate project-owned quality capabilities.
+It does not ship test suites or replace the frameworks and business assets that make a test
+executable.
+
+The first bundled skill is `e2e`. It connects three layers without merging their ownership:
+
+1. a framework such as case-harness may provide reusable execution and verdict primitives;
+2. each business project owns its real E2E code, cases, adapters, and acceptance criteria;
+3. instructions beside that E2E code preserve project-specific setup, failure, cleanup, and evidence
+   knowledge.
+
+The skill reads the project first, discovers its existing E2E entrypoint, executes it when the
+required target is available, and reports evidence. If no runnable suite exists, it returns
+`no_capability` rather than inventing tests or presenting source review as validation.
+
+## Install
+
+Add the devloop marketplace once, then install the plugin.
+
+Claude Code:
+
+```text
+/plugin marketplace add https://github.com/compforge/devloop.git
+/plugin install quality@devloop
+```
+
+Codex:
+
+```bash
+codex plugin marketplace add https://github.com/compforge/devloop.git
+codex plugin add quality@devloop
+```
+
+Start a new session after installation so the agent can discover the bundled skill.
+
+## Use
+
+Ask for the outcome instead of naming a framework command:
+
+```text
+Run this project's E2E tests and report the result.
+Check whether the test environment passes E2E.
+Find and execute the existing E2E capability; tell me if the project has none.
+```
+
+The skill only triggers and evaluates existing project-owned E2E tests. It does not create tests,
+change assertions, fix failures, deploy environments, or turn unavailable coverage into a pass.
diff --git a/quality/skills/e2e/SKILL.md b/quality/skills/e2e/SKILL.md
@@ -0,0 +1,102 @@
+---
+name: e2e
+description: Discover, run, and interpret an existing project's end-to-end tests. Use when the user asks to run, check, inspect, or report E2E tests; validate a project or deployed environment end to end; or determine whether a project has an executable E2E capability. Follow project-owned E2E code and operating knowledge, use case-harness primitives only when the project already integrates them, and never invent missing tests.
+---
+
+# E2E Quality
+
+Evaluate the project's existing E2E capability. Trigger and interpret tests; do not author tests,
+fix failures, or substitute source review for execution evidence.
+
+## Keep the three owners separate
+
+1. Treat case-harness or another framework as infrastructure that may provide runners, lifecycle,
+   assertions, evidence, and verdict primitives.
+2. Treat the business project's E2E code, cases, fixtures, adapters, and acceptance criteria as the
+   executable verification asset.
+3. Treat instructions and runbooks beside that E2E code as the source of project-specific operating
+   knowledge, including environment setup, known failure modes, retries, cleanup, and evidence paths.
+
+Do not assume that installing or importing case-harness means the project has a runnable E2E suite.
+Do not bypass a project wrapper to invoke a lower-level framework directly.
+
+## Understand the project
+
+1. Resolve the repository or project root from the user's target.
+2. Follow the project's AGENTS.md discovery rules, then read its README.
+3. Establish what the project does, which user or API boundary its E2E suite exercises, and which
+   revision and environment the requested run should evaluate.
+
+Do not require a diff or a dedicated quality declaration. Use the repository's current facts.
+
+## Discover the E2E capability
+
+Search the repository before running anything:
+
+- Locate likely E2E, acceptance, integration, system-test, or product-test directories.
+- Inspect nearby AGENTS.md, README, runbooks, Makefiles, package scripts, configuration, fixtures,
+  and recent result artifacts.
+- Search for documented E2E commands and framework entrypoints, including project-local wrappers
+  around case-harness, Playwright, Cypress, pytest, or other runners.
+- Prefer the business-facing suite and its canonical command when multiple lower-level runners
+  exist.
+
+For the selected E2E directory, read its local documentation and configuration before execution.
+Project-local knowledge takes precedence over generic framework habits.
+
+Classify discovery as one of:
+
+- `ready`: an executable suite, target, and required inputs are available;
+- `blocked`: the suite exists but a required environment, credential, dependency, service, device,
+  or target is unavailable;
+- `no_capability`: no executable project-owned E2E suite can be found.
+
+Framework code without business tests is `no_capability`, not `ready`. If more than one plausible
+suite remains and repository evidence cannot select one, report the ambiguity instead of guessing.
+
+## Execute the project-owned entrypoint
+
+1. Use the command documented by the project or exposed through its canonical Make/package/CLI
+   entrypoint.
+2. Preserve the requested revision and environment. Do not deploy, install new dependencies, create
+   credentials, or switch to a different target unless the user explicitly authorizes it.
+3. Respect project-declared timeouts, concurrency, cleanup, and retry policy. Do not retry a failure
+   merely to obtain a green run; only apply retries already defined by the suite or its runbook.
+4. Capture the exact command, exit status, native verdict, logs, screenshots, traces, reports, and
+   other evidence paths produced by the suite.
+
+Do not edit source, cases, assertions, configuration, or expected results during the assessment.
+When execution would have material impact on a shared or production environment, stop for the
+required authorization before running it.
+
+## Interpret the evidence
+
+Use the suite's native result semantics. When case-harness artifacts exist, consume their Run,
+Verdict, checks, and evidence without flattening `skipped` or `error` into pass.
+
+Distinguish these outcomes:
+
+- `passed`: the selected suite completed and its declared checks passed;
+- `failed`: the suite completed and a business assertion or acceptance condition failed;
+- `error`: the runner, setup, cleanup, environment, or evidence collection failed;
+- `blocked`: execution could not start or complete because a prerequisite was unavailable;
+- `no_capability`: no runnable project-owned E2E suite was discovered.
+
+Inspect available failure artifacts to explain what failed, but do not diagnose beyond the evidence
+or turn a plausible cause into a confirmed root cause.
+
+## Report and preserve learnings
+
+Report:
+
+- the project, revision, environment, and E2E boundary evaluated;
+- the E2E directory and canonical entrypoint discovered;
+- the exact capability and command executed;
+- the outcome and the evidence supporting it;
+- failed checks, execution errors, blocked prerequisites, and untested or unknown areas;
+- reusable project-specific operating knowledge observed during the run.
+
+Do not silently modify the project's E2E documentation while assessing it. When a repeatable setup
+issue, workaround, cleanup rule, or evidence location should be retained, propose a concise update
+beside the project's E2E code and apply it only when the user asks. Promote a lesson into this skill
+only after it proves reusable across projects.
diff --git a/quality/skills/e2e/agents/openai.yaml b/quality/skills/e2e/agents/openai.yaml
@@ -0,0 +1,4 @@
+interface:
+  display_name: "E2E Quality"
+  short_description: "Run and interpret project-owned E2E tests"
+  default_prompt: "Use $e2e to discover, run, and report the existing project E2E tests."