The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to ingest and process untrusted data from the open web.
Ingestion points: Raw search snippets from web_search_exa and full webpage content from web_fetch_exa (referenced in references/searching.md and references/extraction.md) are ingested into the agent context for analysis.
Boundary markers: The instructions in SKILL.md and references/extraction.md guide the agent to extract specific structured data into predefined schemas, which helps isolate relevant data from potential injection content, though explicit delimiters for raw content are not mandated.
Capability inventory: The skill uses web search and retrieval tools to gather data and synthesizes it into reports or structured files in the ./exa-results/ directory as described in SKILL.md.
Sanitization: Results are filtered and deduplicated based on relevance and quality criteria (defined in references/filtering.md and references/source-quality.md) before being presented to the user.

exa-search