Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →GitHub security alerts command center -- triage Dependabot, code scanning, and secret scanning alerts entirely from the editor. Bypasses the color-dependent, focus-trapping security UI that is largely inaccessible to screen readers.
.claude/skills/community-access-security-dashboard/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-22 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 42% | 0% |
Derived from .claude/agents/security-dashboard.md. Treat platform-specific tool names or delegation instructions as Codex equivalents.
Shared instructions
Skills: github-workflow-standards, github-scanning
You are the Security Dashboard. You give screen reader users and keyboard-only users full control over GitHub's security features — Dependabot alerts, code scanning results, and secret scanning alerts — whose web UI uses color-coded severity badges, focus-trapping dismissal modals, and visually-overlaid code annotations that are largely inaccessible to assistive technology.
GitHub's security dashboards present severe accessibility barriers:
This agent bypasses all of that by working directly through the GitHub REST API.
dependabot.yml.gh api user.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,715 | 20,771 | +169% | 1 | 1 | 0% | 699 | 1,180 | +69% | 0 | 0 | — |
case-02 | fail→fail | 15,397 | 6,723 | -56% | 1 | 1 | 0% | 835 | 1,134 | +36% | 0 | 0 | — |
case-03 | fail→fail | 16,021 | 6,214 | -61% | 1 | 1 | 0% | 2,890 | 1,086 | -62% | 0 | 0 | — |
case-04 | pass→pass | 9,075 | 12,243 | +35% | 1 | 1 | 0% | 1,630 | 1,758 | +8% | 0 | 0 | — |
case-05 | pass→pass | 7,366 | 6,132 | -17% | 1 | 1 | 0% | 1,290 | 1,826 | +42% | 0 | 0 | — |
case-06 | pass→pass | 10,526 | 3,557 | -66% | 1 | 1 | 0% | 1,525 | 1,388 | -9% | 0 | 0 | — |
case-07 | pass→pass | 6,054 | 9,141 | +51% | 1 | 1 | 0% | 1,067 | 1,741 | +63% | 0 | 0 | — |
case-08 | pass→pass | 10,038 | 5,748 | -43% | 1 | 1 | 0% | 1,910 | 1,848 | -3% | 0 | 0 | — |
case-09 | pass→pass | 4,117 | 3,597 | -13% | 1 | 1 | 0% | 624 | 1,268 | +103% | 0 | 0 | — |
case-10 | pass→pass | 4,925 | 5,611 | +14% | 1 | 1 | 0% | 750 | 1,776 | +137% | 0 | 0 | — |
case-11 | pass→pass | 2,801 | 4,283 | +53% | 1 | 1 | 0% | 454 | 1,589 | +250% | 0 | 0 | — |
case-12 | pass→pass | 9,264 | 11,071 | +20% | 1 | 1 | 0% | 1,480 | 2,527 | +71% | 0 | 0 | — |
case-13 | fail→pass | 7,342 | 7,400 | +1% | 1 | 1 | 0% | 1,277 | 2,066 | +62% | 0 | 0 | — |
case-14 | pass→pass | 10,616 | 7,134 | -33% | 1 | 1 | 0% | 1,962 | 2,169 | +11% | 0 | 0 | — |
case-15 | fail→pass | 16,076 | 12,436 | -23% | 1 | 1 | 0% | 2,673 | 2,843 | +6% | 0 | 0 | — |
case-16 | pass→pass | 9,764 | 5,110 | -48% | 1 | 1 | 0% | 804 | 1,743 | +117% | 0 | 0 | — |
case-17 | pass→pass | 3,858 | 3,572 | -7% | 1 | 1 | 0% | 591 | 1,245 | +111% | 0 | 0 | — |
case-18 | pass→pass | 8,959 | 4,917 | -45% | 1 | 1 | 0% | 1,580 | 1,635 | +3% | 0 | 0 | — |
case-19 | pass→pass | 9,099 | 3,940 | -57% | 1 | 1 | 0% | 1,593 | 1,516 | -5% | 0 | 0 | — |
case-20 | pass→pass | 10,529 | 6,152 | -42% | 1 | 1 | 0% | 1,626 | 1,913 | +18% | 0 | 0 | — |
case-21 | pass→pass | 6,427 | 5,947 | -7% | 1 | 1 | 0% | 1,021 | 1,788 | +75% | 0 | 0 | — |
case-22 | fail→pass | 14,168 | 9,641 | -32% | 1 | 1 | 0% | 2,297 | 2,567 | +12% | 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 19 counted toward the lift figure. The other 3 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 +14 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.