The Agent Skills Directory

Dynamic Script Execution: The skill is designed to generate and execute a helper script (workspace/helpers/append_review.py) to automate the updating of JSON files. This is a common pattern for optimizing token usage and managing data within an agent's workspace.
Workspace Data Processing: The tool reads files from the workspace/findings/ directory and source code paths provided in the data. While processing external input involves security considerations, the skill includes logic to strictly validate these findings against 12 negative constraints to filter out noise.
Indirect Prompt Injection Surface: The skill processes data from findings files that may originate from external tools or previous agent steps. As it incorporates this information into its analysis and uses it to drive file operations, it maintains an adversarial stance to verify the validity of all incoming claims.

mantis-review