adversarial-beam
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates as a static review gate for Elixir source code. It does not include any executable scripts, binary dependencies, or network-enabled tools.
- [PROMPT_INJECTION]: The skill implements a 'blind reviewer' pattern using an isolated subagent. The reviewer prompt template in
references/reviewer-prompt.mdincludes specific instructions to ignore 'intentions stated in comments' in the target code, which effectively mitigates indirect prompt injection risks from the files being analyzed. The subagent is provided with no conversation history, further isolating the execution context. - [DATA_EXFILTRATION]: No data exfiltration patterns were detected. The skill reads repository configuration (e.g.,
mix.exs) and source files to perform its review, but it does not transmit this data to any external or untrusted domains. References point to official Erlang, Elixir, and reputable community documentation. - [COMMAND_EXECUTION]: The skill does not perform any shell command execution or dynamic code evaluation. It relies entirely on text analysis and logical evaluation of the review target against established rules.
- [CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or sensitive environment variable patterns were found in the skill's files.
Audit Metadata