The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze untrusted data from various external sources, including product reviews, social media comments, Reddit, and online forums.
Ingestion points: External data is sourced via tools or manual input as described in SKILL.md and references/voice-of-customer.md to identify 'voice of customer' patterns.
Boundary markers: The instructions do not define clear delimiters or provide the agent with guidance to ignore malicious instructions that might be embedded in the mined content.
Capability inventory: The skill performs file writing (audience.md) and refers to the reddit-marketing skill/tool for subsequent tasks.
Sanitization: No sanitization or filtering mechanisms are prescribed for the external data before it enters the agent's context.

audience-research