Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →リポジトリ内の悪用可能なバウンティ対象のセキュリティ問題を発見します。ノイズの多いローカルのみの発見ではなく、実際のレポートに適格なリモートから到達可能な脆弱性に焦点を当てます。
.claude/skills/affaan-m-security-bounty-hunter/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 12% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 31% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 19% | 0% |
当目标是针对负责任披露或赏金提交的实际漏洞发现，而非广泛的实践审查时使用此方法。
优先关注远程可达、用户可控的攻击路径，并剔除平台通常判定为信息性或超出范围的模式。
以下是持续具有影响力的漏洞类型：
| 模式 | CWE | 典型影响 | | --- | --- | --- | | 通过用户可控URL的SSRF | CWE-918 | 内网访问、云元数据窃取 | | 中间件或API防护中的认证绕过 | CWE-287 | 未授权账户或数据访问 | | 远程反序列化或上传至RCE路径 | CWE-502 | 代码执行 | | 可达端点中的SQL注入 | CWE-89 | 数据泄露、认证绕过、数据破坏 | | 请求处理程序中的命令注入 | CWE-78 | 代码执行 | | 文件服务路径中的路径遍历 | CWE-22 | 任意文件读取或写入 | | 自动触发的XSS | CWE-79 | 会话窃取、管理员权限沦陷 |
除非项目另有说明，以下通常属于低信号或超出赏金范围：
pickle.loads、torch.load 或等效且无远程路径的漏洞eval() 或 exec()shell=Truebashsemgrep --config=auto --severity=ERROR --severity=WARNING --json
然后手动过滤：
markdown## 描述 [漏洞是什么及其重要性] ## 漏洞代码 [文件路径、行号范围及代码片段] ## 概念验证 [最小化可运行的请求或脚本] ## 影响 [攻击者能够实现的目标] ## 受影响版本 [已测试的版本、提交或部署目标]
提交前需确认：
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | pass→pass | 14,433 | 11,003 | -24% | 1 | 1 | 0% | 2,382 | 2,659 | +12% | 0 | 0 | — |
case-05 | pass→pass | 12,655 | 11,395 | -10% | 1 | 1 | 0% | 2,033 | 2,670 | +31% | 0 | 0 | — |
case-01 | fail→fail | 20,433 | 17,043 | -17% | 1 | 1 | 0% | 3,224 | 3,636 | +13% | 0 | 0 | — |
case-02 | pass→pass | 10,724 | 8,000 | -25% | 1 | 1 | 0% | 1,633 | 1,942 | +19% | 0 | 0 | — |
case-03 | pass→pass | 11,276 | 7,095 | -37% | 1 | 1 | 0% | 1,700 | 1,763 | +4% | 0 | 0 | — |
case-04 | pass→pass | 12,276 | 6,395 | -48% | 1 | 1 | 0% | 1,739 | 1,747 | +0% | 0 | 0 | — |
case-07 | pass→pass | 14,322 | 10,381 | -28% | 1 | 1 | 0% | 2,176 | 2,265 | +4% | 0 | 0 | — |
case-08 | pass→pass | 11,260 | 6,540 | -42% | 1 | 1 | 0% | 1,664 | 1,487 | -11% | 0 | 0 | — |
case-09 | pass→pass | 12,382 | 8,459 | -32% | 1 | 1 | 0% | 1,944 | 2,096 | +8% | 0 | 0 | — |
case-10 | pass→pass | 10,517 | 9,408 | -11% | 1 | 1 | 0% | 1,726 | 2,341 | +36% | 0 | 0 | — |
case-15 | pass→pass | 11,510 | 8,981 | -22% | 1 | 1 | 0% | 1,659 | 2,118 | +28% | 0 | 0 | — |
case-11 | pass→pass | 11,193 | 8,026 | -28% | 1 | 1 | 0% | 1,806 | 2,064 | +14% | 0 | 0 | — |
case-12 | pass→pass | 10,323 | 7,128 | -31% | 1 | 1 | 0% | 1,581 | 1,959 | +24% | 0 | 0 | — |
case-13 | pass→pass | 13,084 | 10,314 | -21% | 1 | 1 | 0% | 2,105 | 2,545 | +21% | 0 | 0 | — |
case-14 | pass→pass | 11,588 | 9,580 | -17% | 1 | 1 | 0% | 1,694 | 2,265 | +34% | 0 | 0 | — |
case-16 | pass→pass | 14,844 | 11,653 | -21% | 1 | 1 | 0% | 2,195 | 2,391 | +9% | 0 | 0 | — |
case-17 | pass→pass | 8,009 | 5,064 | -37% | 1 | 1 | 0% | 1,266 | 1,490 | +18% | 0 | 0 | — |
case-18 | pass→pass | 10,229 | 5,504 | -46% | 1 | 1 | 0% | 1,265 | 1,577 | +25% | 0 | 0 | — |
case-19 | fail→pass | 7,233 | 2,714 | -62% | 1 | 1 | 0% | 1,033 | 1,168 | +13% | 0 | 0 | — |
case-20 | fail→pass | 19,967 | 17,946 | -10% | 1 | 1 | 0% | 1,029 | 1,441 | +40% | 0 | 0 | — |
case-21 | pass→pass | 14,342 | 9,062 | -37% | 1 | 1 | 0% | 2,199 | 2,174 | -1% | 0 | 0 | — |
case-22 | pass→pass | 4,750 | 3,381 | -29% | 1 | 1 | 0% | 801 | 1,445 | +80% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.