pentest-ctf-crypto
Warn
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATION
Full Analysis
- COMMAND_EXECUTION (HIGH): The function
execute_command(command)inreferences/tools.mdallows for the execution of arbitrary shell commands. This provides an unconstrained execution environment that could be used for persistence, privilege escalation, or data exfiltration if the agent's logic is subverted. - REMOTE_CODE_EXECUTION (MEDIUM): The
sagemath_execute(script_content)tool allows for the execution of arbitrary SageMath scripts. Because SageMath is Python-based, this allows for general-purpose code execution within the environment. - DATA_EXFILTRATION (LOW): Multiple tools, including
padding_oracle_attackandjwt_analyzer, accept external URLs as input. This creates a risk of Server-Side Request Forgery (SSRF) or out-of-band exfiltration of processed data to attacker-controlled servers. - INDIRECT_PROMPT_INJECTION (LOW): The skill lacks security boundaries for processing untrusted data. * Ingestion points: Ciphertext inputs, JWT tokens, hash files, and URL response data. * Boundary markers: None identified. * Capability inventory: Access to shell commands, script execution, and local file reading. * Sanitization: No sanitization or input validation logic is present in the provided skill files.
Audit Metadata