The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The insight_synthesizer.py script processes research observations from an external JSON file. Because these observations typically contain verbatim notes or quotes from participants, they represent an untrusted data source that is interpolated into the agent's context during the synthesis process.
Ingestion points: scripts/insight_synthesizer.py reads user-provided JSON data via the --input argument.
Boundary markers: The skill does not implement delimiters or safety instructions to distinguish participant notes from agent instructions within the synthesis output.
Capability inventory: The skill includes local scripts for configuration management and research planning but does not possess high-risk capabilities like arbitrary shell execution or remote network access.
Sanitization: No sanitization or validation of the text content within observation notes is performed before processing.

product-research