business-logic-attack
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The documentation in
references/logic-flaw-patterns.mdincludes shell command examples for performing security testing, such as usingffuffor OTP brute-forcing,tesseractfor captcha OCR, andcurlorwebsocatfor testing rate limits and WebSocket vulnerabilities. - [INDIRECT_PROMPT_INJECTION]: The skill establishes a methodology for the agent to ingest and analyze untrusted data from external web application APIs, including payment statuses, coupon codes, and transaction responses.
- Ingestion points: Untrusted external data sources such as HTTP response bodies, API status codes, and redirected URLs analyzed in
SKILL.mdandreferences/logic-flaw-patterns.md. - Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing data from these external endpoints.
- Capability inventory: The skill methodology includes capabilities for network requests (
curl,websocat), local tool execution (ffuf,tesseract), and data manipulation. - Sanitization: There is no evidence of sanitization or strict schema validation for the external API data being processed by the agent.
Audit Metadata