The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources via web_search and web_fetch to build competitive battle cards. This creates a theoretical surface for indirect prompt injection if a competitor's website or review platform contains malicious instructions.
Ingestion points: Competitive intelligence gathered from external URLs and review platforms (G2, LinkedIn, company blogs) via web_search and web_fetch as defined in the Research Methodology section of SKILL.md and Rule 4 of operating-principles.md.
Boundary markers: The skill employs explicit Assumption: prefixes for all internal guesses and uses [User to add: <description>] placeholders to ensure clear separation between agent-generated analysis and required human input.
Capability inventory: The skill is restricted to information retrieval tools (web_search, web_fetch). There are no capabilities for file-system modification, subprocess execution, or outbound network calls to non-standard domains beyond general research tools.
Sanitization: Factual claims are subject to a strict 4-tier source hierarchy (Rule 4), requiring validation from official or reputable third-party sources (Gartner, SEC filings, etc.) before they are considered verified.

competitive-intelligence