The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it retrieves and processes untrusted data from external sources to map narratives.
Ingestion points: The skill utilizes scripts/connectors/firecrawl.py for live web scraping, scripts/connectors/tavily.py for search results, and scripts/connectors/wayback.py for historical archives.
Boundary markers: The instructions include a specific security guardrail directing the agent to treat every scraped page, archived snapshot, or search result as untrusted input and to explicitly ignore any instructions embedded within that content.
Capability inventory: The skill is authorized to write resulting maps to the memory/narrative/category-narrative-mapper/ directory and propose new data to memory/events/claims.ndjson through an internal registry-events.py script.
Sanitization: Although the instructions refer to a SECURITY.md document for handling untrusted data, the skill itself does not specify the precise validation or sanitization routines used before the scraped data is interpolated into the agent's context.

category-narrative-mapper