Home / ccoalm / ccl-skills · GitHub

AGENTS.md@specs/064-eight-patterns-efficacy/evidenceA

AGENTS.md@specs/064-eight-patterns-efficacy/evidence from ccoalm/ccl-skills is an AGENTS.md file: the standing instructions coding agents read in that repo.

Install it, pinned

mdr add ccoalm/ccl-skills/specs/064-eight-patterns-efficacy/evidence/AGENTS.md@git:20260828.130039c

Pins git:20260828.130039c in mdr.lock. Pin the fingerprint instead.

Audit grade: A, 14 of 14 checks passed
Kind: AGENTS.md file
Repository: ccoalm/ccl-skills
Path: specs/064-eight-patterns-efficacy/evidence/AGENTS.md
Versions recorded: 1
Latest version: git:20260828.130039c, committed Aug 28, 2026
Size: 6,150 bytes
Repository stars: 6
License: Apache-2.0

Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.

What the file says

# 064 evidence — 测量设计与边界（引用任何数字前先读）

## 探针设计（probe-064-patterns.rb）

- 形态：023 同款 body-as-prompt 差分。`claude --print --tools "" --model claude-haiku-4-5`，中性 tmp 目录，180s 超时，10 轮/任务，错误轮计 fail 且留痕。
- 两臂：base = 当前 SKILL.md Workflow 段（slice-base.md，sha256 见结果 JSON）；head = 改后 Workflow 段 + 快速选用决策表 + 写完测试走查清单（模拟指针被跟随后可见的面）。
- 判分：关键词合约（机械谓词），跑正式臂前已过 `--selftest`（对已知好/坏样例各能绿能红——oracle 可失败性已证）。

## 这些数字不是什么

1. **不是 merge gate**（eval/AGENTS.md advisory 契约）。
2. **不证明"agent 会跟随指针"**——head 臂直接嵌入了指针指向的内容；真实 agent 需要自己决定读 reference，那是另一个（live-repo）探针形态，本轮未做。head−base 的 delta 只支持「firing point 处可见面变化会改变输出」。
3. **关键词合约 ≠ 语义质量**。谓词判的是结构特征（命名了 smell 类、出现工厂/参数化结构），不是测试代码整体好坏。
4. **单 provider 单 model**。023 先例：Claude 臂的 delta 在 codex 复跑中未复现；本轮结论限定在所用 provider/model，跨 provider 不外推。
5. **污染面**：两臂同环境同注入，差分有效性成立；但 `claude --print` 可能加载用户全局 CLAUDE.md，绝对分可能受染。跑后对 ans_head 抽查 grep 全局指令特征词（CodeGraph / context-mode 等），结果记于下方。

## 判分器校准记录（r1 废弃轮）

第一次 base 臂（probe-064-base-r1-discarded.json，T1 2/10 / T2 0/10 / T3 0/10）判读后**废弃**，原因与处置：

- T1 假红：模型以意译命名问题类（「条件跳过」「混合多个关注点」），初版正则只认近术语。→ 放宽同义词类，并把该意译输出钉进 `--selftest` good[1] 作回归 pin。
…

Read the whole file at its exact version.

How to install

The exact contents, by fingerprint

mdr add ccoalm/ccl-skills/specs/064-eight-patterns-efficacy/evidence/AGENTS.md@sha256:36e41a0cdcaab5e2

The line at the top of this page pins the label git:20260828.130039c, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.

Badge

[![mdr](https://markdownregistry.com/badge/art_hnj4jpy2wcn45ldr.svg)](https://markdownregistry.com/a/art_hnj4jpy2wcn45ldr)

4 badge image fetches in 30 days. What the badge shows.

Versions

The registry has recorded 1 version of this file, committed on Aug 28, 2026.

version	committed	commit	size	audit
git:20260828.130039c latest	2026-08-28	130039c	6,150 B	A	view

Audit of the latest version

A 14 of 14 checks passed. Fixed automatic checks with no AI involved, so the same checks give the same answer every run.

pass: Size between 200 bytes and 200 KB (6150 bytes)
pass: No zero-width or bidi control characters
pass: No instruction hidden inside an HTML comment
pass: No link to an exfiltration or paste host
pass: No credential-shaped string
pass: No instruction to send local credentials anywhere
pass: No text hidden with inline styles
pass: No prompt-injection phrasing
pass: No curl or wget piped into a shell
pass: No recursive delete of root, home or parent
pass: No instruction to read or print local credentials
pass: No base64 blob over 200 characters
pass: No link to a raw IP address
pass: No script tag

Source

GitHub

ccoalm/ccl-skills · 6 stars · license Apache-2.0 · pushed 2026-11-05 · branch main

API

GET https://markdownregistry.com/api/v1/artifacts/art_hnj4jpy2wcn45ldr
GET https://markdownregistry.com/api/v1/resolve?ref=ccoalm/ccl-skills/specs/064-eight-patterns-efficacy/evidence/AGENTS.md
GET https://markdownregistry.com/api/v1/blob/36e41a0cdcaab5e2235913aae7c81f5c8b89f2a0cf935f03f7e2c2920716377d

Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.

Guides: how to write a good AGENTS.md · AGENTS.md vs CLAUDE.md · which agent file does what.