The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and install required CLI tools via a shell script from the official domain https://claude.ai/install.sh piped directly to bash. This is a standard installation procedure for the service.
[EXTERNAL_DOWNLOADS]: Fetches documentation and project resources from established official sources including github.com/anthropics, code.claude.com, and claude.ai.
[COMMAND_EXECUTION]: Demonstrates the use of the Bash tool for agent automation, explicitly recommending the use of pattern-matching permission rules (e.g., allowedTools "Bash(git status)") to restrict executable commands to a safe subset.
[PROMPT_INJECTION]: The skill identifies and addresses an indirect prompt injection surface by providing architectural patterns for agents that process untrusted external data.
Ingestion points: Processes codebase content, file data, and external inputs via the Read and Grep tools.
Boundary markers: Promotes the use of structured JSON schemas and specific system prompts to enforce constraints on model outputs.
Capability inventory: Uses tools with high capabilities including Bash (shell access), Edit (file writing), and Read (file access).
Sanitization: Implements deterministic control flow via HookMatcher to validate tool inputs and block destructive operations (e.g., rm -rf) before they are executed.

agent-builder