skill-comment-insights
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated content from
.txt,.json, and.csvfiles. While the current implementation primarily uses rule-based extraction and sentiment analysis, the resulting 'representative comments' and 'demand examples' are intended to be fed into other downstream tasks or skills. Without proper sanitization, malicious instructions embedded in comments could influence subsequent agent actions. - Ingestion points:
scripts/comment_insights.pyvia_load_commentsfunction which reads external files. - Boundary markers: None present in the instructions for downstream usage.
- Capability inventory: The skill itself has file-write capabilities for reports and terminal output.
- Sanitization: Basic whitespace stripping and encoding replacement are performed, but no content-level sanitization against instruction injection is implemented.
- [COMMAND_EXECUTION]: The skill provides instructions to run a Python script from the shell. This is a standard execution pattern for this type of tool and does not involve elevated privileges or remote script execution.
Audit Metadata