The Agent Skills Directory

[PROMPT_INJECTION]: A static analysis alert for instruction override patterns was identified in the security documentation. Analysis confirms this is a false positive; the file discusses prompt injection as a threat model for event-driven integrations and provides mitigation strategies for developers. The skill also provides legitimate guidance on using emphasis keywords like "IMPORTANT" to improve instruction following.
[COMMAND_EXECUTION]: The documentation describes the functionality of tools like Claude Code and the Agent SDK, which include capabilities for executing shell commands. It includes robust security advice, discussing the trade-offs of autonomous execution flags and providing code examples for defensive "hooks" that block access to sensitive system paths or destructive commands.
[EXTERNAL_DOWNLOADS]: The skill provides official installation instructions for various SDKs and CLI tools using standard package managers (npm, pip, Homebrew) and references official release assets from the "anthropics" GitHub organization.
[SAFE]: No evidence of obfuscation, unauthorized data exfiltration, or malicious persistence was found. The skill serves as an educational and operational guide for building and managing Claude-based agents.

claude-expert