The Agent Skills Directory

[SAFE]: The skill does not contain any evidence of hardcoded credentials, malicious persistence mechanisms, or unauthorized network operations. It uses standard markdown formatting and clear instructional logic.
[SAFE]: References to external services like SparkToro and Reddit are for manual research or well-known information gathering purposes and do not involve remote code execution or automated downloads.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including customer transcripts, survey responses, and community posts (Reddit, G2, etc.) in both Mode 1 and Mode 2. It lacks explicit instructions for the agent to use boundary markers or to ignore instructions that might be maliciously embedded within these external sources (Indirect Prompt Injection).
Ingestion points: Processes raw transcripts, survey results, and forum posts identified in SKILL.md and references/source-guides.md.
Boundary markers: None provided in the instructions.
Capability inventory: The skill includes instructions to synthesize data and hand off to other skills (e.g., copywriting, page-cro), which could propagate injected instructions.
Sanitization: No sanitization or validation steps are defined for the external data being processed.

customer-research