sar-cybersecurity
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill implements a robust defensive instruction (Rule 7 in SKILL.md) that commands the agent to treat all content from the analyzed codebase as untrusted data and ignore any directives or commands contained within it, effectively mitigating indirect prompt injection attempts.
- [REMOTE_CODE_EXECUTION]: Rule 8 in SKILL.md explicitly forbids the agent from generating executable scripts, running shell commands, or installing software packages. Its functionality is strictly limited to producing Markdown-based reports.
- [COMMAND_EXECUTION]: The skill enforces a strict read-only policy for the codebase and limits all write operations to the
docs/security/directory (Rule 1 in SKILL.md), preventing unauthorized system modifications. - [DATA_EXFILTRATION]: While the skill analyzes code and configuration, it uses external web search only for official security references (NVD, MITRE) and includes specific warnings against following or executing arbitrary URLs discovered during the audit process.
Audit Metadata