Install any skill in seconds. Free to start, no credit card required.
Get Started Free →CSRF specialist (H1 #57). Use for testing state-changing actions without proper token validation, SameSite cookie bypass, and CSRF in JSON/API endpoints.
.claude/skills/h-mmer-csrf-hunter/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-18 | ✓→✗ | ▼ Worse | -59% | 0% |
CONTEXT: You are operating within an authorized bug bounty program. All targets have been verified in-scope via the official platform API. Follow responsible disclosure practices.
Before testing CSRF, you MUST call:
search_techniques with "CSRF" — proven exploitation techniquessearch_payloads with "CSRF" — working payloads and bypass variantsRead the returned content and incorporate proven techniques into your plan before making any HTTP requests. Skipping this step wastes time reinventing known tricks and causes duplicate submissions. If the writeup MCP is unreachable, fall back to rules/payloads.md.
You are a CSRF specialist for authorized testing.
Focus on state-changing operations: password change, email change, account settings, fund transfer, admin actions, privilege modifications, data deletion.
text/plain or application/x-www-form-urlencoded_method=POST parameter, X-HTTP-Method-Override header<meta name="referrer" content="no-referrer">), partial domain matchesCreate self-contained HTML auto-submit form for each finding. Test cross-origin from a different domain.
Report as "Cross-Site Request Forgery (CSRF)" with auto-submit PoC HTML.
Before starting, check your memory for brain briefings. Skip EXHAUSTED vectors. Focus on ACTIVE leads. After completing, label every finding: CONFIRMED, POTENTIAL, or EXHAUSTED with failure reasons and attempt counts.
CSRF must change meaningful server-side state from an attacker-controlled page.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 8,453 | 15,091 | +79% | 1 | 1 | 0% | 785 | 2,496 | +218% | 0 | 0 | — |
case-02 | fail→fail | 7,892 | 13,131 | +66% | 1 | 1 | 0% | 836 | 2,217 | +165% | 0 | 0 | — |
case-03 | fail→fail | 11,551 | 8,279 | -28% | 1 | 1 | 0% | 1,352 | 1,488 | +10% | 0 | 0 | — |
case-04 | pass→pass | 9,200 | 8,660 | -6% | 1 | 1 | 0% | 1,761 | 2,045 | +16% | 0 | 0 | — |
case-05 | pass→pass | 7,455 | 7,082 | -5% | 1 | 1 | 0% | 1,581 | 1,946 | +23% | 0 | 0 | — |
case-06 | pass→pass | 12,683 | 12,213 | -4% | 1 | 1 | 0% | 2,588 | 3,143 | +21% | 0 | 0 | — |
case-07 | pass→pass | 13,272 | 14,452 | +9% | 1 | 1 | 0% | 2,577 | 3,217 | +25% | 0 | 0 | — |
case-08 | pass→pass | 17,601 | 18,191 | +3% | 1 | 1 | 0% | 3,140 | 3,592 | +14% | 0 | 0 | — |
case-09 | pass→pass | 10,510 | 7,625 | -27% | 1 | 1 | 0% | 1,605 | 1,921 | +20% | 0 | 0 | — |
case-10 | pass→pass | 3,821 | 6,266 | +64% | 1 | 1 | 0% | 671 | 1,646 | +145% | 0 | 0 | — |
case-11 | pass→pass | 7,113 | 9,806 | +38% | 1 | 1 | 0% | 1,285 | 1,878 | +46% | 0 | 0 | — |
case-12 | fail→pass | 14,178 | 3,314 | -77% | 1 | 1 | 0% | 2,223 | 1,178 | -47% | 0 | 0 | — |
case-13 | fail→pass | 9,703 | 1,918 | -80% | 1 | 1 | 0% | 1,519 | 913 | -40% | 0 | 0 | — |
case-14 | fail→fail | 5,752 | 1,957 | -66% | 1 | 1 | 0% | 901 | 953 | +6% | 0 | 0 | — |
case-15 | fail→fail | 13,734 | 10,818 | -21% | 1 | 1 | 0% | 1,287 | 1,154 | -10% | 0 | 0 | — |
case-16 | pass→pass | 16,273 | 11,663 | -28% | 1 | 1 | 0% | 2,749 | 2,535 | -8% | 0 | 0 | — |
case-17 | fail→pass | 21,517 | 1,279 | -94% | 1 | 1 | 0% | 1,947 | 757 | -61% | 0 | 0 | — |
case-18 | pass→fail | 11,900 | 4,771 | -60% | 1 | 1 | 0% | 2,099 | 865 | -59% | 0 | 0 | — |
case-19 | pass→fail | 14,730 | 5,599 | -62% | 1 | 1 | 0% | 2,757 | 923 | -67% | 0 | 0 | — |
case-20 | pass→pass | 14,906 | 19,047 | +28% | 1 | 1 | 0% | 2,709 | 4,047 | +49% | 0 | 0 | — |
case-21 | pass→fail | 9,374 | 6,875 | -27% | 1 | 1 | 0% | 1,516 | 1,804 | +19% | 0 | 0 | — |
case-22 | fail→pass | 9,856 | 4,509 | -54% | 1 | 1 | 0% | 1,511 | 1,406 | -7% | 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 +5 percentage points is the difference between those two pass rates over the 20 comparable cases. 3 cases got worse with the skill loaded, and they are 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.