The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface. The skill reads and displays data from external files that could contain malicious instructions.
Ingestion points: The extract_frontmatter function in scripts/skill_toggle.py reads SKILL.md files from other skill directories in ~/.claude/skills/.
Boundary markers: Extracted skill metadata is displayed to the agent in a plain text table without protective delimiters or instructions to ignore embedded commands.
Capability inventory: The skill uses the Bash tool to execute a Python script that can move directories and modify the user's ~/.claude/settings.json configuration file.
Sanitization: The script uses a strict alphanumeric regex to prevent path traversal via skill names, though it does not filter extracted descriptions for instructional content.
[COMMAND_EXECUTION]: Local Script Execution. The skill uses the Bash tool to execute its primary logic contained in scripts/skill_toggle.py, which performs filesystem operations and updates configuration files.

skill-toggle