From 5cd30bac2b32c98b7efd4e1996b3517b11d4330c Mon Sep 17 00:00:00 2001
From: Jesse Zhang <101612750+Nagi-ovo@users.noreply.github.com>
Date: Wed, 24 Jun 2026 20:18:08 +0100
Subject: [PATCH] docs: require running the harness as it ships (no extra
 skills)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Address fairness concern from #24: the same harness can load completely
different skills / plugins / MCP / custom system prompts, which shifts the
output as much as swapping the harness and breaks Agent comparability.

- README.md / README.en.md: add a bullet to the Model + Harness section
  stating the harness is evaluated in its default, out-of-the-box form.
- AGENTS.md: add a "Run the harness as it ships" rule — no extra skills
  beyond what the harness bundles; disclose any deviation in `note`.

Enforcement stays honor-system + the PR generation screenshot.

Closes #24

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 AGENTS.md    | 12 ++++++++++++
 README.en.md |  1 +
 README.md    |  1 +
 3 files changed, 14 insertions(+)

diff --git a/AGENTS.md b/AGENTS.md
index 4e4d617..a749f8b 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -71,6 +71,18 @@ Every run needs `note.zh` + `note.en` documenting provenance: which harness,
 which effort, one-shot or fixed. If you applied any fix, say where the bug was
 and what you changed. **No emojis anywhere** (hard rule). zh first.
 
+### Run the harness as it ships / 用 harness 开箱即用的默认形态
+
+Use the harness in its **default, out-of-the-box** configuration: no extra
+skills, plugins, MCP servers or custom system prompts beyond what it bundles.
+These shift the output as much as swapping the harness does, so loading them
+makes the run no longer comparable to others under the same harness. If you used
+any, it is a different setup — say so explicitly in `note`, otherwise it does
+not belong on the board.
+（用 harness 的**默认形态**：不要额外加载 harness 自带以外的 skill / 插件 /
+MCP / 自定义系统提示——它们对产出的影响不亚于换 harness，会让这条 run 失去与
+同一 harness 下其它条目的可比性。确实用了就在 `note` 里写清楚，否则不该上榜。）
+
 ## Before you open the PR / 提交前
 
 ```bash
diff --git a/README.en.md b/README.en.md
index ba25e12..840ca53 100644
--- a/README.en.md
+++ b/README.en.md
@@ -21,6 +21,7 @@ One-shot LLM eval cases by NAGI STUDIO: same prompt, different model x harness x
 - **Model**: the weights and inference engine itself — GPT-5.5, Gemini 3.1 Pro, Claude Fable 5. Reasoning effort is a capability dial of the model itself: the same weights at high vs low effort reason at depths that feel like two different models, so effort belongs to the "model" dimension alongside the weights — judging a model while ignoring its effort isn't fair play. This bench defaults to max and labels the effort of every combination.
 - **Harness**: the product/scaffolding wrapping the model — Codex CLI, Cursor, AntiGravity, the Claude web app. It controls tool calls, system prompts, context management and continuation strategy, and often shapes the outcome as much as the model does.
 - The unit of evaluation here is therefore an **Agent**: the same model (including its effort) under a different harness is a different agent and a separate entry (GPT-5.5 Pro and GPT-5.5 running in Codex CLI at xhigh are two agents from the same model family).
+- **The harness is evaluated as it ships, out of the box**: do not load skills / plugins / MCP servers / custom system prompts beyond what it bundles by default — they shift the output as much as swapping the harness does and break comparability between agents. If you did use any, it is no longer the same agent; disclose it in the `note`.
 
 ## Registry
 
diff --git a/README.md b/README.md
index 488b7b4..e402ee9 100644
--- a/README.md
+++ b/README.md
@@ -21,6 +21,7 @@ NAGI STUDIO 的 LLM 测评案例集：同一段提示词，不同「模型 × Ha
 - **模型（Model）**：权重与推理引擎本身，如 GPT-5.5、Gemini 3.1 Pro、Claude Fable 5。思考配额（effort）是模型自身的能力旋钮——同一份权重在高/低思考强度下的推理深度判若两个模型，因此它和权重同属「模型」这一维度，脱离 effort 谈模型表现并不公允；本仓库默认拉满（Max），并在每个组合里如实标注。
 - **Harness（运行环境）**：包裹模型的产品/脚手架，如 Codex CLI、Cursor、AntiGravity、Claude 网页版——它决定工具调用、系统提示词、上下文管理与续写策略，对最终产出的影响往往不亚于模型本身。
 - 因此本仓库的测评单位是一个 **Agent**：同一个模型（含其思考配额）换一个 Harness，就是另一个 Agent，记为不同条目（例如 GPT-5.5 Pro 与跑在 Codex CLI 里的 GPT-5.5（xhigh）是同一模型家族的两个 Agent）。
+- **测的是 Harness 开箱即用的默认形态**：请勿额外加载 Harness 自带以外的 skill / 插件 / MCP / 自定义系统提示——它们对产出的影响不亚于换 Harness，会破坏 Agent 之间的可比性。确实用了，就不是同一套 Agent，必须在 `note` 里如实说明。
 
 ## 已测组合 / Registry