The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external log files (state.json, events.jsonl) generated from previous runs. These logs can contain untrusted data from various tool outputs. There is a potential risk that malicious instructions contained within these logs could influence the analysis or the suggested improvements if not properly sanitized.
[DYNAMIC_EXECUTION]: The process involves generating modifications (patches) for SKILL.md files. While this is the intended purpose of the tool, it represents the creation of new instructions or logic that should be reviewed by a human before application.

skill-improver