The Agent Skills Directory

[COMMAND_EXECUTION]: Employs shell utilities like find, ls, du, and git hash-object to perform integrity audits on the .claude/ configuration directory and installed plugins.
[COMMAND_EXECUTION]: Includes the capability to delete identified junk files and temporary archives using the rm command during automated or guided fix operations.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection due to its handling of external project data.
Ingestion points: Reads content from configuration files such as .claude/.gitignore, settings.json, settings.local.json, and SKILL.md manifests within skill subdirectories.
Boundary markers: Does not employ explicit delimiters or system instructions to ignore potential commands embedded within the analyzed configuration files.
Capability inventory: Accesses tools for file deletion (rm), git repository manipulation, and project-wide file reading.
Sanitization: No evidence of input validation or sanitization is present for data ingested from the project files before it is processed or reported.

claude-health