The Agent Skills Directory

[PROMPT_INJECTION]: The skill establishes a research workflow that ingests data from untrusted external environments, creating an attack surface for indirect prompt injection.
Ingestion points: As outlined in SKILL.md and references/research.md, the agent is instructed to research user pain and market signals on platforms such as Reddit, Quora, LinkedIn, X, and GitHub.
Boundary markers: The instructions do not define boundary markers or delimiters to isolate retrieved data, nor do they instruct the agent to ignore instructions that might be embedded in the researched text.
Capability inventory: The agent uses this external information to generate artifacts (pitch decks, decision memos, MVP boundaries) and steer product decisions; these outputs could be manipulated by adversarial data within the research sources.
Sanitization: There are no explicit requirements for the agent to sanitize or filter the content retrieved from external sources before it is processed for synthesis and artifact generation.

product-ideation