The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates and examples for bash-based hooks that execute shell commands to parse configuration files. It specifically demonstrates the use of sed, grep, awk, and jq for data processing, and tmux send-keys for inter-process coordination between agents.
[INDIRECT_PROMPT_INJECTION]: The skill implements a pattern where the agent reads instructions and configuration from project-local files (.claude/*.local.md) to drive its behavior, creating a surface for indirect injection if these files are influenced by untrusted data.
Ingestion points: The skill reads settings and prompts from .claude/plugin-name.local.md files within the project directory.
Boundary markers: Uses standard YAML frontmatter delimiters (---) to separate structured configuration from the markdown prompt body.
Capability inventory: The patterns described include shell command execution (via hooks), file system modifications (atomic updates using mv and cat), and agent loop control (using jq to block exit and provide new prompts).
Sanitization: The skill includes a 'Security Considerations' section explicitly recommending input sanitization for settings files and file path validation to prevent traversal attacks.

plugin-settings