multi-source-search
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external web sources, which serves as an attack surface for instructions hidden in search results or web pages.
- Ingestion points: Search and page-reading capabilities are used to retrieve external content (SKILL.md).
- Boundary markers: Explicit instructions are provided to treat retrieved content as untrusted evidence and ignore embedded instructions (SKILL.md).
- Capability inventory: The skill utilizes local command execution to run a validation script (scripts/validate_report.py) and involves network operations via search/retrieval tools as well as file read/write operations for report generation.
- Sanitization: External content is processed into a structured JSON format and validated using a schema-enforcement script (references/report-schema.md, scripts/validate_report.py).
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script to validate the research report.
- Evidence: The command
python3 scripts/validate_report.py research-report.jsonis described in SKILL.md. - Note: This is a read-only operation on a bundled script intended for data integrity verification and structural analysis.
Audit Metadata