The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses benign shell commands for environment discovery and project context. It executes git branch and git status to identify relevant markdown files and uses a local script detect-caveman.sh to check for the presence of the caveman plugin via the claude CLI tool.
[EXTERNAL_DOWNLOADS]: The skill documentation references external tools including markdownlint-cli2 and a caveman plugin. It provides instructions for manual installation via standard package managers (npm) and does not perform any silent or unauthorized downloads at runtime.
[SAFE]: The skill implements security best practices for file operations, such as using mktemp -d for isolated temporary workspaces and trap commands to ensure temporary files are deleted upon exit.
[SAFE]: The skill includes a robust adversarial verification mechanism where a fresh-context subagent audits changes for semantic loss, preventing the model from grading its own edits and ensuring integrity of the compressed content.

compress