The Agent Skills Directory

[SAFE]: The skill serves as a documentation reference card for various text compression modes and related tools. All instructions are consistent with providing a user-facing help menu.
[DATA_EXPOSURE]: The skill mentions reading configuration from ~/.config/caveman/config.json and environment variables. This is standard practice for managing user preferences and does not involve exfiltration of sensitive data.
[COMMAND_EXECUTION]: The skill describes a post-execution workflow where the agent summarizes its actions and writes them to a specific project directory (llmwiki/wiki/draft/cave/). These file system operations are constrained to a project-specific path and are intended for record-keeping.
[INDIRECT_PROMPT_INJECTION]: The skill includes a template for generating output reports based on recent agent actions. While this processes data from previous tasks, the use of structured Markdown templates minimizes the risk of the agent following instructions embedded in that data. The risk is assessed as safe within the context of standard logging.

caveman-help