The Agent Skills Directory

SAFE (SAFE): The skill acts as a documentation and template repository for plugin development. It promotes security-conscious practices, such as restricting tool access via allowed-tools and using portable environment variables like ${CLAUDE_PLUGIN_ROOT} to prevent path traversal or breakage across environments.
COMMAND_EXECUTION (LOW): The skill instructs the agent to make scripts executable using chmod +x and to configure hooks that execute shell commands.
Evidence: Found in SKILL.md, best-practices/common-mistakes.md, and schemas/plugin-manifest.md.
Context: This behavior is fundamental to the 'Plugin Authoring' use case, as executable scripts and hooks are required components of the plugins being developed.
EXTERNAL_DOWNLOADS (SAFE): The skill references official Anthropic documentation.
Evidence: URL https://docs.anthropic.com/en/docs/agents-and-tools/agent-skills/skill-authoring-best-practices in SKILL.md.
Context: The domain is a Trusted External Source, and the links are for reference purposes only.
INDIRECT_PROMPT_INJECTION (SAFE): The skill has a defined surface for processing repository files but includes strong guardrails.
Ingestion points: Reads plugin.json, marketplace.json, and markdown files in component directories.
Boundary markers: Implements a 'Diagnose-Propose-Execute' flow and utilizes a secondary plugin-reviewer agent for audits.
Capability inventory: File system access (Read, Grep, Glob) and execution of development-specific tools (/plugin-development:*).
Sanitization: Mandates the use of /plugin-development:validate for schema and structural integrity checks.

plugin-authoring