The Agent Skills Directory

Indirect Prompt Injection Surface: The skill documents a pattern for reading configuration from files in the .claude/ directory. Ingestion points: Project-local markdown files are parsed by the agent or hooks. Boundary markers: The skill uses YAML frontmatter delimiters (---) to separate configuration from instructions. Capability inventory: This pattern allows for file system operations, inter-process communication using tmux, and control over agent decision-making. Sanitization: The documentation specifically recommends the use of jq --arg to safely handle content when constructing JSON responses for the agent.\n- Command Execution through Shell Scripting: The skill provides examples of using standard Unix utilities like sed, grep, and awk to manage configuration. These are used in a controlled manner for parsing and updating local files, following best practices like atomic updates with temporary files.\n- Path Traversal Security: The documentation includes clear logic to validate file paths provided in configuration, specifically checking for .. to prevent traversal attacks. This is a critical practice for any plugin that handles project files based on user settings.\n- Local State Protection: The skill emphasizes security by advising users to set restrictive file permissions and to ensure that local configuration files are not committed to version control, which helps maintain the integrity of project-specific state.

Plugin Settings