The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user queries and provide relevant information from local reference files. This creates a surface for indirect prompt injection, as the agent must interpret untrusted user input to select the appropriate documentation context.
Ingestion points: User questions regarding tool setup, commands, and features as defined in the activation examples in SKILL.md.
Boundary markers: The skill relies on standard instruction-following rather than explicit delimiters for user input in the provided examples.
Capability inventory: The documentation references capabilities of the Claude Code tool, including filesystem modifications (write, edit) and shell command execution (bash).
Sanitization: The instructions emphasize user review of scripts and plugins as a security measure.
[EXTERNAL_DOWNLOADS]: The documentation describes installation procedures for the Claude Code CLI and its extensions (plugins and MCP servers) from external sources. These references target well-known package registries and trusted organizations.
Evidence: References to @anthropic-ai/claude-code (NPM), claude-code (PyPI), and various @modelcontextprotocol/ packages in references/getting-started.md and references/mcp-integration.md.
[COMMAND_EXECUTION]: The skill contains a large catalog of shell commands and slash commands (e.g., /cook, /fix:hard, /git:cm, /deploy) intended for user guidance. These commands demonstrate how to interact with the Claude Code terminal interface and perform development workflows.
Evidence: Detailed usage examples and command descriptions across references/slash-commands.md, references/best-practices.md, and references/troubleshooting.md.

claude-code