The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads project configuration files (CLAUDE.md) and rule files (.md files in .claude/rules/) to calculate their size and verify content preservation. This access is local and necessary for its documented auditing function.
[EXTERNAL_DOWNLOADS]: The skill's documentation and scripts reference the author's GitHub repository (github.com/sangrokjung/claude-forge) as the source for the methodology. No automatic network requests or script downloads were found in the provided files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests markdown content from rule files. Because the logic is restricted to byte-counting and exact-string verification for auditing purposes, the risk of interpreted data as malicious instructions is low.
[COMMAND_EXECUTION]: The skill includes utilities that perform standard file system operations like reading file metadata and contents. These actions are performed by explicitly invoked scripts or configured shell hooks, following standard development workflows.

harness-diet