The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides utility scripts (parse-frontmatter.sh, validate-settings.sh) and example hooks (read-settings-hook.sh) that utilize standard Unix tools such as sed, awk, grep, and jq for file processing and configuration management.
[PROMPT_INJECTION]: Documentation describes a pattern where agent behavior is influenced by project-local configuration files (.claude/*.local.md), creating a surface for indirect prompt injection.
Ingestion points: Project-local settings files stored in the .claude/ directory (documented in SKILL.md and examples/example-settings.md).
Boundary markers: The pattern explicitly utilizes YAML frontmatter delimiters (---) to separate structured configuration data from markdown instruction blocks.
Capability inventory: The documented pattern enables the reading and writing of project files and the execution of shell-based hooks that can block tool operations based on configuration (documented in SKILL.md and references/real-world-examples.md).
Sanitization: The skill provides proactive security guidance, including implementation examples for path traversal validation (..), sensitive file blocking (.env, secret), input escaping in shell scripts, and the use of safe JSON construction methods (jq --arg) to prevent injection during agent communication.

Plugin Settings