agent
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a set of instructional guidelines and technical references for managing AI subagents. No malicious code, exfiltration patterns, or obfuscation were detected.
- [PROMPT_INJECTION]: The skill contains explicit warnings and design patterns to mitigate prompt injection, such as using 'boundary markers' in delegation contracts and advising against 'reflexive verification' where an agent checks its own work.
- [DATA_EXFILTRATION]: Mentions of sensitive file paths (e.g.,
~/.aws/credentials,.env) are used strictly as examples for teaching agents what to protect or exclude from subagent scope, rather than being accessed by the skill itself. - [EXTERNAL_DOWNLOADS]: The skill references several external domains (e.g.,
code.claude.com,artificialanalysis.ai,coderabbit.ai) solely for the purpose of citing benchmarks and official documentation. These are plain-text references and do not involve remote code execution.
Audit Metadata