The Agent Skills Directory

[SAFE]: The skill serves as educational documentation and a source of code templates for developers. It does not perform any autonomous actions or include malicious instructions.- [COMMAND_EXECUTION]: The skill includes numerous Bash script examples intended for plugin hooks. These scripts utilize standard POSIX utilities (sed, awk, grep, cat, mv) and common CLI tools (jq, tmux, yq) to parse local configuration files and facilitate inter-process communication. All examples are standard for the described use case of plugin state management.- [PROMPT_INJECTION]: The skill describes a pattern where the markdown body of a .local.md file is used as context or a prompt for the agent. This introduces a surface for indirect prompt injection, which is addressed by the skill's own security recommendations. Evidence chain: 1. Ingestion points: .claude/*.local.md files (SKILL.md, real-world-examples.md). 2. Boundary markers: YAML frontmatter delimiters (---) are used to separate config from body content. 3. Capability inventory: File operations (mv, cat), text processing (sed, awk, grep), and agent output (jq, tmux). 4. Sanitization: The skill explicitly provides sanitization patterns for user input and path traversal validation in its 'Security Considerations' section.

plugin-settings