nice-try-gpt
Harden CTF Challenges Against LLM Shortcuts
Challenge authors need to reduce shallow LLM pattern matching without making fair challenges frustrating. This skill reproduces the intended solve, selects a small resistance change, and verifies the result end to end.
使用我的智慧代理安裝
將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。
Review the Skillstore skill "nice-try-gpt" from https://skillstore.io/skills/aleff-github-nice-try-gpt.md and its manifest at https://skillstore.io/api/skills/aleff-github-nice-try-gpt/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。
Agent 可讀資源
當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。
測試它
正在使用「nice-try-gpt」。 An authorized web challenge exposes an object identifier in a static response, making the intended authorization weakness immediately obvious.
預期結果:
Shortcut identified: static object disclosure. Recommended move: reveal the identifier through ordinary runtime activity, adding one observation without changing the vulnerability.
正在使用「nice-try-gpt」。 A proposed transformation adds a five-step state machine and requires an unrelated protocol.
預期結果:
Rejected by the Human Cost Gate. It adds a new exploit primitive and materially increases player effort, so the original challenge should remain unchanged.
正在使用「nice-try-gpt」。 The transformed challenge starts normally, preserves the weakness, removes the static cue, and passes an end-to-end solve.
預期結果:
TRANSFORMED PASS. The report records the baseline, selected resistance, human-cost estimate, post-change difficulty, verification result, and changed files.
安全審計
安全All 17 static detections are false positives. The command detections match Markdown backticks and status labels, while the reconnaissance detections describe CTF review criteria rather than host discovery or system probing.
風險因素
分享與引用此報告
分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。
複製報告連結
https://skillstore.io/skills/aleff-github-nice-try-gpt/audits/2?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/aleff-github-nice-try-gpt?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/aleff-github-nice-try-gpt?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/aleff-github-nice-try-gpt/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/aleff-github-nice-try-gpt.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>學術引用 (APA · BibTeX · CFF)
APA 引用
aleff-github. (2026). nice-try-gpt security audit report (audit version 2) [Author version 0.2.0]. Skillstore. https://skillstore.io/skills/aleff-github-nice-try-gpt/audits/2BibTeX 引用
@techreport{aleff-github-aleff-github-nice-try-gpt-2026,
author = {aleff-github},
title = {nice-try-gpt security audit report (audit version 2)},
institution = {Skillstore},
year = {2026},
number = {2},
url = {https://skillstore.io/skills/aleff-github-nice-try-gpt/audits/2},
note = {Author version 0.2.0}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "nice-try-gpt security audit report (audit version 2)"
version: "0.2.0"
type: report
authors:
- name: "aleff-github"
date-released: "2026-09-20"
url: "https://skillstore.io/skills/aleff-github-nice-try-gpt/audits/2"
identifiers:
- type: other
value: "skillstore:aleff-github-nice-try-gpt:audit:2"
description: "Skillstore immutable audit report identifier"
Skillstore 評分
評分依據 證據置信度: 中你可以打造什麼
Review a Training Challenge
Reproduce a lab solve, identify the fastest shallow shortcut, and propose a small change that preserves the lesson.
Validate a Challenge Update
Check that a transformed challenge still starts, exposes the intended weakness, and produces the expected result.
Improve LLM-Assisted Practice
Reduce obvious model shortcuts while keeping the expected human knowledge, actions, and difficulty band stable.
試試這些提示
Review this authorized CTF challenge. Record its vulnerability class, intended solve path, required knowledge, difficulty band, and clean-start procedure. Do not modify files.
Using the verified baseline evidence, identify the single cheapest LLM shortcut. Explain which player-facing clue enables it and why the shortcut does not represent the intended learning objective.
Choose the smallest resistance move that reduces the identified shortcut. Check the same vulnerability class, learning objective, prerequisite knowledge, and human-cost gate before proposing the change.
Verify the transformed challenge from a clean state. Check startup, legitimate behavior, intended vulnerability, shortcut reduction, runtime success, flag isolation, and independent-solver status. Produce the required final report.
最佳實務
- Start with a clean, reproducible baseline before proposing any change.
- Prefer one evidence-based resistance move with one to three additional meaningful actions.
- Verify the intended solve, flag isolation, and normal application behavior after every change.
避免
- Do not hide flags behind randomness, unstable behavior, or undiscoverable values.
- Do not stack decoys, brute force, CAPTCHA, token flooding, or unrelated trivia.
- Do not claim LLM resistance without an isolated fresh-solver check or an explicit NOT TESTED result.
常見問題
Who should use this skill?
Does it create new vulnerabilities?
Can it test production systems?
How many changes should it make?
How does it verify success?
Does self-review prove model resistance?
開發者詳細資訊
作者
aleff-github授權
GPL-3.0-only
作者版本
v0.2.0
Skillstore 修訂
r2
版本提示
可安裝內容已變更，但作者沒有更新聲明的版本號。
引用
9f1a7e8b681745b04f88358f3cf08da94782cac9
維護新鮮度
2026/9/20
使用量
0 次下載 · 0 次瀏覽
檔案結構