The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process external data through a multi-stage pipeline, creating an indirect prompt injection surface where malicious instructions in input data could influence agent behavior.
Ingestion points: In scripts/pipeline_template.py, the stage_acquire function is a placeholder for fetching data from external sources (APIs, databases, web scraping).
Boundary markers: The PROMPT_TEMPLATE in scripts/pipeline_template.py uses markdown headers (e.g., '# Content to Analyze') to delimit input, but it does not include explicit instructions for the LLM to ignore potentially malicious instructions embedded within that content.
Capability inventory: The provided template script performs file system operations (read/write) within the local data/ and output/ directories. It does not currently implement active network or shell execution, though the methodology discusses such tools.
Sanitization: There is no evidence of input sanitization or escaping in the generate_prompt function before data is interpolated into the prompt template.

project-development