The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and processing content from external web searches, creating a surface for indirect prompt injection attacks.
Ingestion points: Web content retrieved via host search tools and third-party providers such as Tavily, Exa, and Scholar as defined in SKILL.md and references/sandbase-api-map.md.
Boundary markers: The skill instructions explicitly warn the agent to treat all retrieved content as untrusted evidence and to ignore any prompt injections or operational instructions embedded in those results.
Capability inventory: The skill utilizes network search capabilities and file read/write operations for generating and validating research reports.
Sanitization: The provided scripts/validate_report.py script performs structural and URL-normalization checks on the output but does not sanitize the text content of the claims themselves.
[SAFE]: No malicious patterns were detected. The skill implements security best practices, including explicit privacy warnings for external queries, guidance on keeping API keys out of logs, and a read-only default workflow to prevent unintended external modifications. The included Python script for validation is safe and performs no network operations.

multi-source-search