[Skills] Standardize the benchmark integration template and one-click verification workflow (based on the env/prmeval standard).

.gitignore

@@ -32,3 +32,4 @@ AGENTS.md
 
 # Local Harbor dataset tooling
 /env/harbor/generate_vulhub_dataset.py
+.zcode/

README.md

@@ -51,33 +51,55 @@ https://github.com/user-attachments/assets/4c551b27-ce4d-4fc8-8df6-d6dc8100cc88
 
 ## <a id="agent-skill"></a>🧩 Agent Skill Quick Start
 
-This repository includes a lightweight Agent skill that helps agents use SAfactory through the standard workflows:
+This repository includes an Agent skill for template-based environment integration, local contract tests, optional Docker/RJob evaluation, and GRPO/RL workflows:
 
 ```text
 skills/safactory-workflows/SKILL.md
 ```
 
-It covers three common requests:
+### <a id="benchmark-onboarding-prompt"></a>Benchmark Onboarding Prompt
 
-- onboard a new benchmark or custom environment into SAfactory;
-- run Docker-mode evaluation for a selected environment;
-- start GRPO / RL training for a selected environment.
+Provide the benchmark source, one-row dataset schema, 1–2 representative cases, native single-case command, and native output format. Specify Docker or RJob deployment when known. **Integration alone does not require evaluation, a score/reward definition, a running Gateway, or an internal cluster.** Only provide scoring details when evaluation is requested.
 
-When working with an Agent, use prompts such as:
+The Agent copies fixed runner/config templates, fills the environment hooks, and runs local contract tests with an owned mock model endpoint. Native dependencies must be available locally or represented by explicit test fixtures. Live deployment is a separate check when image/data/model/runtime access is available; a helper starts and stops Gateway without a second terminal. RJob live runs additionally require an existing cluster configuration and a cluster-reachable Gateway URL.
 
-```text
-Use skills/safactory-workflows to help me onboard this benchmark into SAfactory.
-```
+<details>
+<summary>Expand to get the Benchmark Onboarding Prompt</summary>
 
 ```text
-Use the safactory-workflows skill to run geo3k evaluation in Docker mode.
+Use skills/safactory-workflows to onboard this benchmark into SAfactory.
+
+[Goal]
+- evaluation: [disabled (default, integration only) / enabled]
+- target deployment mode: [docker / rjob / decide after adapter inspection]
+- validation: [local contract first / also run live deployment when available]
+
+[Benchmark]
+- environment name: ____________________
+- source or checkout path/repository: ____________________
+- dataset path and one-row schema: ____________________
+- 1–2 case IDs/rows: ____________________
+- native single-case command, or README file/section: ____________________
+- Docker image, if available: ____________________
+- native result/output path and format: ____________________
+
+[Only if evaluation is enabled]
+- native score field/file, range, meaning, and pass condition: ____________________
+
+Start from the fixed templates based on docs/guides/custom-environment.md.
+Keep protocol handling in runner.py and fill adapter.py with single-case row
+mapping, Gateway model routing, native command execution, and output collection.
+Do not reimplement native benchmark solving/scoring logic.
+Only add rule_evaluator.py and --enable-evaluation if evaluation is enabled.
+Run local contract checks without requiring Gateway or RJob cluster setup.
+Report fixtures used, native output mapping, and which checks actually passed;
+keep local contract, live deployment, and evaluation results distinct.
+Ask only for missing information needed by the next dependent step.
 ```
 
-```text
-Use the safactory-workflows skill to start GRPO training for my_env.
-```
+</details>
 
-The skill does not replace the docs. It guides the Agent to read `docs/guides/`, `docs/reference/`, and the root README as needed, while using the standard `env/geo3k/` environment as the reference implementation. If your Agent supports local skill discovery, add `skills/safactory-workflows/` to its skill search path; otherwise mention this path explicitly in the request.
+See [Custom Environments](docs/guides/custom-environment.md) and the skill's [integration reference](skills/safactory-workflows/references/environment-integration.md) for the templates and Docker/RJob deployment settings. Verify any environment with one command — `python skills/safactory-workflows/scripts/check_environment.py --env env/<name>` — which runs static consistency checks, a contract smoke test against an owned mock endpoint, and an optional live stage, labeling each failure with the side that owns it. If your Agent cannot discover local skills automatically, include `skills/safactory-workflows/` explicitly in the prompt.
 
 ## <a id="quick-start"></a>🚀 Quick Start
 

README_CN.md

@@ -51,33 +51,53 @@ https://github.com/user-attachments/assets/4c551b27-ce4d-4fc8-8df6-d6dc8100cc88
 
 ## <a id="agent-skill"></a>🧩 Agent Skill 快速上手
 
-仓库内置了一个轻量 Agent skill，用于帮助 Agent 按标准 workflow 使用 SAfactory：
+仓库内置 Agent skill，支持从固定模板接入环境、本地契约测试、可选的 Docker/RJob 评测，以及 GRPO/RL workflow：
 
 ```text
 skills/safactory-workflows/SKILL.md
 ```
 
-它覆盖三类高频请求：
+### <a id="benchmark-onboarding-prompt"></a>Benchmark 接入 Prompt
 
-- 接入新的 benchmark 或自定义环境到 SAfactory；
-- 用 Docker 模式运行指定环境的测评；
-- 启动指定环境的 GRPO / RL 训练。
+请提供 benchmark 源码、单条 dataset row 格式、1–2 个代表性 case、原生单 case 命令和输出格式。已知时注明 Docker 或 RJob 部署模式。**仅接入不要求评测、score/reward 定义、已启动的 Gateway 或内部集群。** 只有需要评测时才提供评分信息。
 
-使用 Agent 时，可以直接这样提问：
+Agent 会复制固定 runner/config 模板，填写环境 hook，并使用自动管理的 mock 模型端点执行本地契约测试。原生依赖需要本地可用或有明确的测试 fixture。镜像、数据、模型和 runtime 可用后再做真实部署验证；helper 负责 Gateway 启停，无需另开终端。RJob 真实运行还需要已有集群配置及集群可达的 Gateway URL。
 
-```text
-请使用 skills/safactory-workflows，帮我把这个 benchmark 接入 SAfactory。
-```
+<details>
+<summary>展开获取接入 Prompt</summary>
 
 ```text
-请使用 safactory-workflows skill，用 Docker 模式跑 geo3k 测评。
+请使用 skills/safactory-workflows，将以下 benchmark 接入 SAfactory。
+
+【本次目标】
+- evaluation: [disabled（默认，仅接入）/ enabled]
+- 目标部署模式: [docker / rjob / 检查 adapter 后再确定]
+- 验证范围: [先做本地契约验证 / 条件具备时也做真实部署验证]
+
+【Benchmark 信息】
+- environment name: ____________________
+- 源码或 checkout 路径/仓库地址: ____________________
+- 数据集路径和单条 row 格式: ____________________
+- 1–2 个 case ID/row: ____________________
+- 原生单 case 命令，或 README 文件/章节: ____________________
+- Docker image（如已有）: ____________________
+- 原生结果输出路径和格式: ____________________
+
+【仅在启用评测时填写】
+- 原生分数字段/文件、范围、含义、通过条件: ____________________
+
+从基于 docs/guides/custom-environment.md 的固定模板开始。
+runner.py 保留协议处理，adapter.py 只填写单 case 的数据映射、Gateway 模型
+配置、原生命令调用和输出收集。不要重写 benchmark 内部的解题/评分逻辑。
+仅在启用评测时添加 rule_evaluator.py 和 --enable-evaluation。
+先运行本地契约检查，不要求手动启动 Gateway 或配置 RJob 集群。
+报告使用的 fixture、原生输出映射及实际通过的检查；区分本地契约、真实部署
+和评测结果。仅询问下一步依赖的缺失信息。
 ```
 
-```text
-请使用 safactory-workflows skill，启动 my_env 环境的 GRPO 训练。
-```
+</details>
 
-该 skill 不替代文档，而是引导 Agent 按需读取 `docs/guides/`、`docs/reference/` 和根 README，并优先参考标准环境 `env/geo3k/`。如果你的 Agent 支持本地 skill 搜索，可以把 `skills/safactory-workflows/` 加入其 skill 搜索路径；否则在请求中显式写出该路径即可。
+固定模板和 Docker/RJob 配置见[自定义环境指南](docs/guides/custom-environment_CN.md)和 skill 的[接入参考](skills/safactory-workflows/references/environment-integration.md)。任意环境可用一条命令完成验证：`python skills/safactory-workflows/scripts/check_environment.py --env env/<name>`，它会依次执行静态一致性检查、基于自有 mock 端点的契约冒烟以及可选的 live 阶段，并为每条失败标注归属方（配置侧 / 环境侧 / 框架侧）。如果 Agent 不支持自动发现本地 skill，请在 prompt 中显式写出 `skills/safactory-workflows/` 路径。
 
 ## <a id="quick-start"></a>🚀 快速开始