The Agent Skills Directory

[DATA_EXFILTRATION]: The enrichment protocol (Step 4b) explicitly instructs the agent to "Send existing brain knowledge as context" to external search tools such as Perplexity, Brave Search, and Exa. This practice aims to fetch delta updates but effectively exposes internal knowledge, which may contain private notes or relationship details, to third-party services.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including social media posts and web scraping (Steps 4b, 4c), to extract "texture" such as beliefs and motivations. This creates an attack surface where malicious instructions embedded in a target's profile or recent posts could influence the agent during processing.
Ingestion points: External search results, social media feeds, and various enrichment APIs (SKILL.md, Step 4).
Boundary markers: While the skill uses markdown structure and source citations, it lacks specific instructions or delimiters to prevent the agent from obeying commands found within the retrieved text.
Capability inventory: The skill uses put_page for file writes, get_page for reads, and search/query for data retrieval.
Sanitization: There is no mention of sanitizing or escaping the content retrieved from external sources before it is written into the local knowledge base.
[EXTERNAL_DOWNLOADS]: The skill relies on multiple third-party APIs for entity enrichment, including Crustdata, Proxycurl, People Data Labs, Crunchbase, PitchBook, and Clearbit. While these are recognized industry services, the skill depends heavily on their availability and data integrity.

enrich