CLAUDE.md@docs/measurements/2026-10-24-s4-inputsA
CLAUDE.md@docs/measurements/2026-10-24-s4-inputs from kopp0510/claude-dd is a CLAUDE.md file: the standing instructions Claude Code reads in that repo.
mdr add kopp0510/claude-dd/docs/measurements/2026-10-24-s4-inputs/CLAUDE.md@git:20261023.7f05f62Pins git:20261023.7f05f62 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 14 of 14 checks passed
- Kind
- CLAUDE.md file
- Repository
- kopp0510/claude-dd
- Path
docs/measurements/2026-10-24-s4-inputs/CLAUDE.md- Versions recorded
- 2
- Latest version
- git:20261023.7f05f62, committed Oct 23, 2026
- First version recorded
- Oct 23, 2026
- Size
- 2,476 bytes
- Repository stars
- 12
- License
- MIT
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
# docs/measurements/2026-10-24-s4-inputs
裁決審查 S4（`docs/designs/2026-10-23-裁決審查-design.md`）量測用的**凍結實驗輸入**：三組（新做法、haiku 對照、舊做法）
拿到的 prompt、共用的段落內容、產生快照與統計的腳本。結果寫在上一層的 `2026-10-23-reviewer-real-sessions.md`「新做法對照」一節。
## 檔案
- `briefs.txt` —— 每段（G2、G3、G4）的範圍與已跑過的驗證；三組逐字共用，**不帶**舊報告派工節的「重點／點名」提示
- `judge.txt` —— 新做法與 haiku 組的裁決者說明（照部署版全域 §3.9 步驟 4，拿掉「審查者報告逐字寫進 scratchpad 檔」那句，因為允許的工具沒有 Write）；`{REVIEWER}` 決定審查者帶不帶 `model: "haiku"`
- `old-reviewer.txt` —— 舊做法給 `code-reviewer` 的說明（舊報告實際的派工寫法：`git show HEAD:`、最多一次重現實驗）
- `parent.txt` —— headless 父 session 的派工指示（`new` 給新做法與 haiku 組、`old` 給舊做法）
- `prepare.py` —— 每一發建一份快照 repo（`git init`、只有 base 與 change 兩個 commit，拿掉 `docs/reviews/` 底下除了 `CLAUDE.md` 的報告），產生 diff 檔與 prompt 檔
- `run.sh` —— 在快照裡跑一發 `claude -p`，stdout 存 `<s4>/logs/<發代號>.jsonl`
- `stats.py` —— 讀父 log 的全部 `result` 與 `~/.claude/projects/<快照路徑編碼>/` 的子代理紀錄，算耗時、工具次數、模型、費用，並列作廢理由（denials、中止、層數、模型、洩題）
- `test_leaky.py` —— `stats.py` 洩題判定 `leaky()` 的負面測試（該抓的、實跑出現過的誤報各幾則）；改 `leaky()` 後跑 `python3 -I test_leaky.py`，全過 exit 0
- 結果（每發一列的 `stats.jsonl`、每個 agent 最後一則文字）在旁邊的 `2026-10-24-s4-outputs/`
## 約束
…Read the whole file at its exact version.
How to install
mdr add kopp0510/claude-dd/docs/measurements/2026-10-24-s4-inputs/CLAUDE.md@sha256:9fb4e4a5a0036458The line at the top of this page pins the label git:20261023.7f05f62, 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.
[](https://markdownregistry.com/a/art_ip72jniottr2z5za)
4 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 2 versions of this file, all committed on Oct 23, 2026.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20261023.7f05f62 latest | 2026-10-23 | 7f05f62 | 2,476 B | A | view · diff |
| git:20261023.972062d | 2026-10-23 | 972062d | 2,270 B | A | view |
Audit of the latest version
- pass: Size between 200 bytes and 200 KB (2476 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
kopp0510/claude-dd · 12 stars · license MIT · pushed 2026-10-30 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_ip72jniottr2z5za GET https://markdownregistry.com/api/v1/resolve?ref=kopp0510/claude-dd/docs/measurements/2026-10-24-s4-inputs/CLAUDE.md GET https://markdownregistry.com/api/v1/blob/9fb4e4a5a0036458836a565f9200f46f9a144c20c7bb90701beaecbe16631dc5
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: what a CLAUDE.md file is · keeping AGENTS.md and CLAUDE.md in sync · which agent file does what.