skills/seqis/openclaw-skills-converted-from-claude-code/agent-comparison-analyzer/Gen Agent Trust Hub
agent-comparison-analyzer
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill contains no executable scripts, obfuscated code, or unauthorized network operations. Its behavior is consistent with its stated purpose of file analysis.
- [PROMPT_INJECTION]: The skill possesses an inherent attack surface for indirect prompt injection given its primary function.
- Ingestion points: Methodology Phase 1 (File Discovery) and Phase 3 (Content Comparison) in
SKILL.mdrequire the agent to read and process untrusted file content from the local filesystem. - Boundary markers: None present; the instructions do not specify delimiters or warnings to ignore instructions found within files being compared.
- Capability inventory: The skill references tools including
Bash,Read,Grep,Glob, andTodoWrite(listed in the Imported Agent Spec). - Sanitization: There are no instructions for sanitizing or validating the content of the files before processing.
- Conclusion: This risk is a known property of file-processing agents and, in this context, does not elevate the verdict as there is no evidence of the agent being directed to automatically execute code extracted from these files.
Audit Metadata