The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from external competitor websites, creating a surface for indirect prompt injection attacks.
Ingestion points: The 'Process' section in Step 2 instructs the agent to use WebFetch on competitor homepages, pricing pages, and comparison pages, and WebSearch for recent news.
Boundary markers: There are no instructions to wrap fetched content in delimiters or to apply 'ignore embedded instructions' warnings when the agent processes the retrieved text.
Capability inventory: The skill has file-write access to knowledge/markets/competitors.md and the output/ directory.
Sanitization: The instructions lack requirements for sanitizing, escaping, or filtering the content retrieved from external sources before it is analyzed and stored in the internal knowledge base.
[DATA_EXFILTRATION]: The skill reads sensitive internal strategy files and subsequently performs network operations, which could be exploited for exfiltration.
Evidence: The skill reads knowledge/markets/positioning.md and knowledge/icp/personas.md to establish context before using WebFetch and WebSearch to access external competitor domains. This read-then-fetch pattern is a potential vector for exposing internal positioning data if the agent's behavior is influenced by adversarial input.

competitor-analyst