exploit-chain-building
Fail
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: CRITICALDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The skill includes a pre-configured payload in 'REFERENCE.md' that uses the 'fetch' API to exfiltrate 'document.cookie' to an external domain ('attacker.example.com'). This domain was flagged as malicious/phishing by automated security scanners.\n- [INDIRECT_PROMPT_INJECTION]: The skill's workflow in 'SKILL.md' accepts user-controlled inputs via 'SECSKILL_INITIAL_FINDING' and 'SECSKILL_TARGET_URL'. These inputs are interpolated directly into bash commands and markdown reports.\n
- Ingestion points: 'SKILL.md' (SECSKILL_INITIAL_FINDING, SECSKILL_TARGET_URL)\n
- Boundary markers: None present.\n
- Capability inventory: 'curl' (network interaction), 'cat' and 'mkdir' (file system access).\n
- Sanitization: None detected.\n- [COMMAND_EXECUTION]: The skill provides instructions and shell snippets in 'SKILL.md' and 'REFERENCE.md' to execute 'curl' commands against sensitive cloud metadata endpoints (169.254.169.254) for the purpose of credential extraction.
Recommendations
- AI detected serious security threats
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata