The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted customer interview transcripts, which creates a surface for indirect prompt injection if a transcript contains adversarial instructions.
Ingestion points: The scripts/interview_synthesizer.py script reads transcript data from JSON files provided via the --input command-line argument.
Boundary markers: The system lacks explicit instructions or delimiters that tell the agent to disregard any natural language instructions that might be embedded within the processed transcript text.
Capability inventory: The skill includes a Python script capable of writing files to the local filesystem (--output).
Sanitization: There is no evidence of sanitization, filtering, or escaping of the transcript content to remove or neutralize potentially malicious instructions before they are clustered into themes or rendered in the final report.

interview-synthesis