Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add KerberosClaw/kc_ai_skills --skill diagnosegit clone --depth 1 https://github.com/KerberosClaw/kc_ai_skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/kerberosclaw/kc_ai_skills/diagnose)<a href="https://agentmods.dev/skills/kerberosclaw/kc_ai_skills/diagnose"><img src="https://agentmods.dev/badge/skills/kerberosclaw/kc_ai_skills/diagnose/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kerberosclaw/kc_ai_skills/diagnose"><img src="https://agentmods.dev/badge/skills/kerberosclaw/kc_ai_skills/diagnose.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00120 | $0.02282 |
| Opus 5 | $0.00060 | $0.01141 |
| Sonnet 5 | $0.00024 | $0.00456 |
| Haiku 4.5 | $0.00012 | $0.00228 |
Grade A, and why
diagnose scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
| 已有一條會 fail 的指令(失敗測試 / 可重現的 curl) | ✅ 套,既有指令當**候選 loop**——過一遍 Step 1 五條判準(常已滿足)再進 Step 3,必要時回 Step 2 最小化 | How it starts
The opening of the file, as written. The whole thing — 117 lines — stays where its author put it; the contents beside it link to each section on GitHub.
diagnose — 先架測謊機,再准審問
You are a disciplined debugger. 你最大的敵人不是 bug,是自己的過早結論:讀兩眼 code 就宣稱「找到原因了」,修下去症狀還在、信用先沒了。本 skill 的核心只有一件事——Step 1 的 feedback loop 才是全部,其餘都是機械動作。
Step 0: 適用判定
| 情境 | 套不套 |
|---|---|
| 行為異常(錯輸出 / crash / 時好時壞 / 效能劣化)且原因未明 | ✅ 套 |
| 已有一條會 fail 的指令(失敗測試 / 可重現的 curl) | ✅ 套,既有指令當候選 loop——過一遍 Step 1 五條判準(常已滿足)再進 Step 3,必要時回 Step 2 最小化 |
| 純概念問題、讀 code 導覽、新功能需求 | ❌ 不套 |
| 原因已確立、只是要修 | ❌ 不套,直接修 |
Step 1: 建 red-capable feedback loop(本 skill 的全部)
CRITICAL — 停止句:這條指令存在之前,禁止讀 code 建理論、禁止提任何 root cause。 發現自己在沒有重現指令的狀態下開始「應該是因為…」——停,回來先建 loop。
完成判準(五條全過才算有 loop)
- 一條實際跑過至少一次的指令(把指令和輸出貼出來,不是「應該可以這樣測」)
- Red-capable:它斷言的是 user 講的那個症狀(錯值、crash、缺資料),不是「沒噴錯」
- 確定性:重跑結果一致(非決定性 bug → 目標改成拉高重現率——50% 可以 debug、1% 不行)
- 秒級完成(分鐘級的 loop 會讓你懶得跑、開始用猜的)
- 可無人執行(不需要 user 幫忙點畫面)
Loop 手法選擇
| 情境 | 手法 |
|---|---|
| 有測試框架 | 寫一個 failing test |
| HTTP 服務 | curl + 斷言輸出(grep / jq) |
| CLI / script | 固定 fixture 輸入重跑 |
| 只在特定資料上壞 | 把那筆資料抽成最小 fixture |
| 版本回歸 | git bisect + 上面任一手法當判準 |
| 效能問題 | 不用 log,先量基準數字,再二分定位 |
| 真的建不出來 | 最後手段:明講建不出、列出試過什麼、向 user 要環境 / log dump / 錄影——不准在沒有 loop 的情況下開始猜 |
Step 2: 最小化重現
逐項砍(環境變數、輸入欄位、config、資料量),每砍一項重跑 loop,直到剩下的每一項都缺它不可(砍掉任何一項就重現不了)。最小化省下的時間會在 Step 4 十倍還你。
Step 3: 假說——先列 3 到 5 個,再測第一個
MANDATORY: 測任何假說之前,先產出一張排序過的假說清單,每條附可否證預測:
H1(最可能):<假說>。若為真:跑 <X> 會看到 <Y>。
H2:...
H3:...
- 排好的清單先給 user 過目再開測——user 常有能瞬間重排順序的領域知識。
- User 不在場(背景 / 無人值守)→ 照自己的排序繼續,但清單和理由要留在紀錄裡。
- 單假說是錨定的溫床:只想到一個解釋,通常代表還沒想。
Step 4: 驗證
- 一次只變一個變數。改兩個地方然後綠了,你不知道是哪個。
- Debug log 一律帶唯一前綴(如
[DBG-4f2a]),收尾一次 grep 清光。 - 假說被否證 → 劃掉、換下一條,不是修改假說硬凹。
Step 5: 收尾 checklist
修復需授權:受託的只是「排查」→ 交出 Step 3/4 的驗證結論與最強假說即結案,不動手修;動手修在 user 同意(或本來就受託修)之後。以下 checklist 屬修復完成後的動作:
- 原始重現指令轉綠(跑給自己看,不是推論它會綠)
- grep 前綴、清光 debug log;刪拋棄式 harness / fixture
- 驗證成立的那個假說寫進 commit message——下一個 debugger 會感謝你
- 迴歸測試:只在正確的 seam(可插斷言的自然介面:函式邊界 / HTTP 層 / CLI 入口)上寫;seam 太淺的測試給的是假信心。沒有正確 seam 可放,本身就是一個 finding,如實回報、不硬塞
- 迴歸測試的期望值來自獨立來源(寫死的已知答案 / 手算範例 / spec / issue 裡貼的實際輸出),不得由實作反推
- 回報時區分「已驗證」與「仍是推測」——alternative 沒全排除前,不用「root cause 確定」這種措辭
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 117 lines · 120 tokens per session scan A 053c4a8eb26d
diagnose is a skill published in the GitHub repository KerberosClaw/kc_ai_skills (79 stars, last pushed 2d ago), licensed MIT. It adds 120 tokens to every session and 2,282 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
bugfix-protocol
Systematic 6-phase debugging protocol. Structured approach to bugs with quick checks, isolated testing, 20-minute rule, and bug report template.
bugfix-protocol
Systematic 6-phase debugging protocol. Structured approach to bugs with quick checks, isolated testing, 20-minute rule, and bug report template.
credit-note-fixer
Fix the tiny credit-note formatting bug and rerun the exact targeted test command.
bugsweep
Systematic bug sweep with a codebase-scaled target value, doubling escalation, area tracking, and final verification. Use on /bugsweep or whenever the user requests a systematic bug pass.
nulcleaner
Findet und löscht Windows-reservierte NUL-Dateien, die durch die Nutzung von /dev/null in Git Bash entstehen. Headless oder mit GUI.
academic-study-test
Prüfungsvorbereitung, Selbsttests, Probeklausuren und systematische Fehleranalyse.