The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the arcjet library directly from the vendor's official GitHub repository (github.com/arcjet/arcjet-py.git) using a specific commit hash for installation.
[COMMAND_EXECUTION]: Provides instructions for the user to execute shell commands, specifically pip install, to set up the environment and install necessary dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input via the Claude Agent SDK, which creates an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via the user_text variable in the query function (found in SKILL.md).
Boundary markers: The skill does not explicitly show boundary markers but focuses on implementing DetectPromptInjection rules to mitigate risks.
Capability inventory: The skill uses query for model interaction and guard_tool to wrap various tool functions, which could include file system or network operations depending on the tools integrated.
Sanitization: The skill demonstrates how to use Arcjet's security rules to detect and block malicious prompts at the UserPromptSubmit hook.

integrate-arcjet-guard-claude-agent-sdk-py