troubleshoot
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves fetching and processing data from external web sources which constitutes an attack surface for indirect prompt injection.
- Ingestion points: Untrusted data enters the agent's context through
WebSearchandWebFetchoperations targeting sites like Stack Overflow, GitHub, and Reddit (referenced inSKILL.mdandprotocols/search-multi-source.md). - Boundary markers: The instructions lack specific delimiters or warnings for the agent to ignore potentially malicious instructions embedded within the fetched web content.
- Capability inventory: Across its scripts, the skill utilizes tools for reading files (
Read), searching file patterns (Glob,Grep), and performing network requests (WebSearch,WebFetch). It also creates and appends to local files in the$PRAXIS_DIRdirectory. - Sanitization: No explicit sanitization, validation, or filtering mechanisms are defined for the content retrieved from external sources before it is analyzed by the model.
Audit Metadata