draft-security-advisory
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted vulnerability reports, creating a surface for indirect prompt injection where malicious instructions could influence the agent's behavior.\n
- Ingestion points: Vulnerability reports enter the agent's context through file paths, pasted text, or GitHub advisory IDs as specified in the
Inputsandargument-hintfields ofSKILL.md.\n - Boundary markers: The instructions do not include boundary markers or explicit guidance for the agent to treat report content as untrusted or to ignore embedded instructions.\n
- Capability inventory: The skill is granted significant capabilities including
Bash(git, gh, and npm),Read,Write,Edit, andAgenttools, enabling it to perform file operations and network requests via the GitHub API.\n - Sanitization: There is no requirement for the agent to sanitize or validate the contents of the report before using it to investigate the codebase or construct command arguments.\n- [COMMAND_EXECUTION]: The skill executes multiple shell commands using the
Bashtool. Several command arguments, such as code snippets forgit log -Sin Phase 3 or advisory IDs forgh apiin Phase 1, are extracted directly from the untrusted vulnerability report, creating a potential for the agent to execute malformed or unintended commands based on manipulated input.
Audit Metadata