The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from a variety of public external sources, which creates a potential surface for indirect prompt injection attacks where malicious instructions could be embedded in user reviews or forum posts.
Ingestion points: Untrusted data enters the agent's context during the 'source sweep' step, where it fetches content from review platforms (G2, Capterra, TrustRadius), app stores (iOS, Android), and community forums (Reddit, practitioner boards) as described in SKILL.md and the provided examples.
Boundary markers: The instructions in SKILL.md require the agent to use verbatim quotes with source URLs and maintain a strict separation of observations from interpretations using 'Fact/Inference/Assumption' labels.
Capability inventory: The skill leverages the agent's web browsing and search capabilities to collect data. The 'Final Step' allows the aggregated data to be passed to other workflows, such as discovery-interview-prep and battle-card-builder.
Sanitization: The skill contains logical safeguards, explicitly stating that every surfaced theme is a 'hypothesis to validate, never a verdict' and requiring source bias notes, though it does not specify technical prompt delimiters for the external content.
[EXTERNAL_DOWNLOADS]: The skill instructions involve fetching data and content from several well-known external platforms and services.
Evidence: SKILL.md directs the agent to sweep mixed voice sources including G2, Capterra, TrustRadius, Reddit, and official app stores.
Note: The skill references a public repository on GitHub (github.com/deanpeters/product-manager-prompts) as the source of its prompt logic and methodology.

voice-of-customer-miner