The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection via external data sources.
Ingestion points: Untrusted data enters the agent's context through WebSearch, mcp__firecrawl__firecrawl_search, and mcp__brave-search__brave_web_search, as well as professional APIs (e.g., Semantic Scholar, ArXiv) as described in references/data-sources.md.
Boundary markers: There are no defined delimiters or instructions provided in the skill files to help the agent distinguish between its core instructions and data retrieved from the web.
Capability inventory: The skill is granted Write and Edit permissions for local file management and WebFetch for network operations.
Sanitization: No evidence of sanitization, filtering, or validation of the retrieved external content is present in the provided implementation patterns.
[DATA_EXFILTRATION]: The skill documentation includes patterns for making requests to external data providers.
Network operations: The reference file references/data-sources.md contains Python implementation examples for fetching person and company data from third-party domains including api.apollo.io, clearbit.com, nubela.co (Proxycurl), and api.peopledatalabs.com.
Context: While these operations are directed towards well-known professional data services and are aligned with the skill's purpose of network reconstruction, they involve network communication with non-whitelisted domains using placeholder API keys.

hr-network-analyst