offensive-crypto-attacks
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill workflow involves ingesting and analyzing untrusted external data, including source code, binaries, and configurations, to perform security audits. This creates an attack surface where instructions embedded in the audited data could attempt to subvert agent behavior.
- Ingestion points: Audited source code, binary files, web bundles, and configuration entrypoints identified during the discovery phase (SKILL.md).
- Boundary markers: There are no explicit instructions to use delimiters or ignore-markers when the agent processes these untrusted inputs.
- Capability inventory: The skill is intended to generate remediation code and execute validation commands, which are sensitive operations that could be misdirected by injected instructions.
- Sanitization: No sanitization or filtering protocols for audited content are specified in the instructions.
- [COMMAND_EXECUTION]: The skill instructs the agent to use tools like
rg(ripgrep) for static analysis andnpm testfor regression verification. While these are standard professional tools, they involve shell-level execution (SKILL.md).
Audit Metadata