captcha-bypass-methodology
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The methodology describes the use of automation and parsing tools, including Selenium, Requests, and Tesseract OCR, to simulate user interaction and solve CAPTCHAs during security assessments.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a process that ingests untrusted data from external web environments (such as CAPTCHA images, API response bodies, and JavaScript files), which creates a potential surface for indirect prompt injection if the agent processes these responses to determine its next action.
- Ingestion points: Target website response headers, body content, and image assets mentioned in Phase 1 and 2 of
SKILL.md. - Boundary markers: None provided in the methodology to separate untrusted web content from agent instructions.
- Capability inventory: Network requests and browser automation capabilities via
requestsandseleniumas described in Phase 2. - Sanitization: The methodology does not specify sanitization or validation procedures for data retrieved from target systems.
Audit Metadata