Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Harden Intent handling, WebView configuration, and FileProvider access in Android apps. Use when securing Intent extras, configuring WebViews, or exposing files via FileProvider; defer manifest export flags and generic Bundle typing to focused Android security guidance.
.claude/skills/hoangnguyen0403-android-legacy-security/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -16% | 0% |
| case-03 | ✓→✓ | = Same ✓ | -12% | 0% |
| case-13 | ✓→✓ | = Same ✓ | -20% | 0% |
android:exported="false" for all internal Activities/Services unless needed for deep links.resolveActivity before starting implicit intents.See hardening examples for manifest and component restrictions.
javaScriptEnabled = false. Use WebViewClient and WebChromeClient to restrict navigation.allowFileAccess and allowFileAccessFromFileURLs to prevent local file theft via XSS.@JavascriptInterface (API 17+), strictly limit exposed API surface.See hardening examples for WebView lockdown patterns.
file:// URIs. Use FileProvider to generate content:// URIs with temporary permissions.EncryptedSharedPreferences for auth tokens and PII. Never use legacy public-read file modes.NetworkSecurityConfig to disable cleartextTrafficPermitted and implement certificate pinning.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 25,225 | 21,639 | -14% | 1 | 1 | 0% | 5,318 | 3,184 | -40% | 0 | 0 | — |
case-02 | fail→pass | 19,133 | 12,859 | -33% | 1 | 1 | 0% | 3,566 | 2,778 | -22% | 0 | 0 | — |
case-03 | pass→pass | 11,458 | 9,343 | -18% | 1 | 1 | 0% | 2,499 | 2,199 | -12% | 0 | 0 | — |
case-13 | pass→pass | 15,341 | 7,953 | -48% | 1 | 1 | 0% | 2,586 | 2,074 | -20% | 0 | 0 | — |
case-04 | pass→pass | 7,483 | 7,152 | -4% | 1 | 1 | 0% | 1,608 | 1,736 | +8% | 0 | 0 | — |
case-05 | fail→fail | 16,549 | 11,875 | -28% | 1 | 1 | 0% | 3,058 | 2,570 | -16% | 0 | 0 | — |
case-06 | pass→pass | 9,466 | 6,552 | -31% | 1 | 1 | 0% | 1,901 | 1,689 | -11% | 0 | 0 | — |
case-07 | fail→pass | 12,694 | 7,987 | -37% | 1 | 1 | 0% | 2,407 | 2,026 | -16% | 0 | 0 | — |
case-18 | pass→pass | 3,442 | 2,721 | -21% | 1 | 1 | 0% | 564 | 760 | +35% | 0 | 0 | — |
case-08 | pass→pass | 16,261 | 11,551 | -29% | 1 | 1 | 0% | 3,189 | 2,680 | -16% | 0 | 0 | — |
case-09 | pass→pass | 9,453 | 5,467 | -42% | 1 | 1 | 0% | 1,661 | 1,529 | -8% | 0 | 0 | — |
case-10 | pass→pass | 10,685 | 8,196 | -23% | 1 | 1 | 0% | 2,015 | 1,955 | -3% | 0 | 0 | — |
case-11 | pass→pass | 13,099 | 9,946 | -24% | 1 | 1 | 0% | 2,457 | 2,202 | -10% | 0 | 0 | — |
case-12 | pass→pass | 16,028 | 9,683 | -40% | 1 | 1 | 0% | 2,875 | 2,150 | -25% | 0 | 0 | — |
case-14 | pass→pass | 14,373 | 8,415 | -41% | 1 | 1 | 0% | 2,884 | 1,985 | -31% | 0 | 0 | — |
case-15 | pass→pass | 8,082 | 7,470 | -8% | 1 | 1 | 0% | 1,586 | 1,791 | +13% | 0 | 0 | — |
case-16 | pass→pass | 6,151 | 6,656 | +8% | 1 | 1 | 0% | 1,199 | 1,686 | +41% | 0 | 0 | — |
case-17 | pass→pass | 7,885 | 3,932 | -50% | 1 | 1 | 0% | 1,162 | 1,049 | -10% | 0 | 0 | — |
case-19 | pass→pass | 6,393 | 4,724 | -26% | 1 | 1 | 0% | 1,349 | 1,301 | -4% | 0 | 0 | — |
case-20 | pass→pass | 8,953 | 8,360 | -7% | 1 | 1 | 0% | 1,860 | 2,060 | +11% | 0 | 0 | — |
case-21 | pass→pass | 7,815 | 9,816 | +26% | 1 | 1 | 0% | 1,557 | 1,638 | +5% | 0 | 0 | — |
case-22 | pass→pass | 5,937 | 4,917 | -17% | 1 | 1 | 0% | 1,137 | 1,253 | +10% | 0 | 0 | — |
case-23 | pass→pass | 10,449 | 7,283 | -30% | 1 | 1 | 0% | 1,989 | 1,789 | -10% | 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. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 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.