The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is primarily focused on ingesting and synthesizing data from untrusted external sources, such as customer interview transcripts, survey results, and online community posts (Reddit, G2, YouTube comments). This creates a surface for indirect prompt injection attacks where an adversary could place malicious instructions within these public sources to manipulate the agent during the analysis process.
Ingestion points: The skill instructions in SKILL.md (Mode 1 and Mode 2) and the platforms detailed in references/source-guides.md (Reddit, G2/Capterra, Hacker News, LinkedIn, Amazon Reviews, YouTube, TikTok) represent the entry points for untrusted data.
Boundary markers: The skill lacks explicit instructions for the agent to use delimiters or ignore potential commands found within the data being analyzed, which increases the likelihood of the agent inadvertently following embedded instructions.
Capability inventory: The agent is expected to generate synthesis reports, personas, and quote banks, which typically requires file-writing capabilities to save the output. It also utilizes search and information retrieval tools to access the digital sources.
Sanitization: There are no provisions in the instructions for sanitizing, filtering, or escaping the content retrieved from external sources before it is processed or synthesized into deliverables.

customer-research