Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review ClawHub or local Skill packages before installation, classify risk, and return a structured security report.
.claude/skills/uvwt-skill-vetter-runtime/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 123% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -43% | 0% |
This Skill provides a security-first review workflow and an optional helper script for inspecting ClawHub slugs or local Skill directories.
Use it before installing unknown skills, when comparing candidate skills, or when checking whether a local Skill has suspicious code, broad permissions, credential access, persistence hooks, or risky network behavior.
The output is a structured JSON report with reviewed files, detected red flags, permission clues, risk level, verdict, and notes.
Skill 本体是本说明文档。确需调用包内辅助脚本时，在 Skill 包根目录使用相对路径执行；运行宿主负责切换到包根目录并把所需变量注入当前子进程。
bashprintf '%s' '{"skill_action":"<动作>"}' | python3 run.py
输入必须是 JSON 对象。写操作仍按本文档中的确认规则执行。
| 动作 | 用途 | |---|---| | status | Check Python runtime and optional ClawHub CLI availability. | | vet-clawhub-slug | Fetch and vet a ClawHub skill by slug without installing it. | | vet-local-path | Vet a local skill folder. By default only AgentDock workspace and installed skill roots are allowed. |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 21,436 | 15,669 | -27% | 1 | 1 | 0% | 3,235 | 674 | -79% | 0 | 0 | — |
case-02 | fail→pass | 18,651 | 25,943 | +39% | 1 | 1 | 0% | 1,577 | 3,512 | +123% | 0 | 0 | — |
case-03 | fail→fail | 8,121 | 23,803 | +193% | 1 | 1 | 0% | 1,235 | 631 | -49% | 0 | 0 | — |
case-04 | fail→pass | 10,413 | 7,642 | -27% | 1 | 1 | 0% | 1,469 | 668 | -55% | 0 | 0 | — |
case-05 | fail→pass | 12,738 | 7,049 | -45% | 1 | 1 | 0% | 2,028 | 617 | -70% | 0 | 0 | — |
case-06 | fail→pass | 11,253 | 10,608 | -6% | 1 | 1 | 0% | 1,091 | 1,179 | +8% | 0 | 0 | — |
case-07 | fail→pass | 7,100 | 13,439 | +89% | 1 | 1 | 0% | 1,087 | 623 | -43% | 0 | 0 | — |
case-08 | fail→pass | 13,979 | 23,354 | +67% | 1 | 1 | 0% | 2,163 | 1,644 | -24% | 0 | 0 | — |
case-09 | fail→pass | 8,199 | 6,899 | -16% | 1 | 1 | 0% | 1,374 | 519 | -62% | 0 | 0 | — |
case-10 | fail→pass | 7,408 | 4,163 | -44% | 1 | 1 | 0% | 1,258 | 706 | -44% | 0 | 0 | — |
case-11 | fail→pass | 15,410 | 13,713 | -11% | 1 | 1 | 0% | 1,825 | 1,032 | -43% | 0 | 0 | — |
case-12 | fail→pass | 6,242 | 5,258 | -16% | 1 | 1 | 0% | 1,065 | 1,064 | -0% | 0 | 0 | — |
case-13 | fail→pass | 14,665 | 5,764 | -61% | 1 | 1 | 0% | 2,365 | 1,296 | -45% | 0 | 0 | — |
case-14 | pass→pass | 17,945 | 5,335 | -70% | 1 | 1 | 0% | 1,978 | 1,019 | -48% | 0 | 0 | — |
case-15 | fail→pass | 13,400 | 6,322 | -53% | 1 | 1 | 0% | 1,917 | 1,063 | -45% | 0 | 0 | — |
case-16 | pass→pass | 18,854 | 12,985 | -31% | 1 | 1 | 0% | 2,186 | 1,489 | -32% | 0 | 0 | — |
case-17 | fail→pass | 13,746 | 7,555 | -45% | 1 | 1 | 0% | 1,358 | 729 | -46% | 0 | 0 | — |
case-18 | pass→pass | 10,236 | 8,101 | -21% | 1 | 1 | 0% | 1,545 | 715 | -54% | 0 | 0 | — |
case-19 | pass→pass | 11,664 | 2,007 | -83% | 1 | 1 | 0% | 973 | 591 | -39% | 0 | 0 | — |
case-20 | pass→pass | 16,582 | 11,970 | -28% | 1 | 1 | 0% | 1,631 | 860 | -47% | 0 | 0 | — |
case-21 | pass→pass | 13,064 | 13,158 | +1% | 1 | 1 | 0% | 2,582 | 2,708 | +5% | 0 | 0 | — |
case-22 | pass→pass | 14,868 | 19,573 | +32% | 1 | 1 | 0% | 3,088 | 3,344 | +8% | 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, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +59 percentage points is the difference between those two pass rates over the 20 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.