The Agent Skills Directory

Command Execution Surface: The skill provides instructions for using the InvokeAgentRuntimeCommand API, which enables the execution of shell commands within an agent's session. While this is a powerful capability intended for development tasks like building and testing, the skill includes explicit warnings for developers to sanitize user-supplied inputs to prevent command injection. It also recommends using restricted IAM policies to limit access to this feature.
Dynamic Code and Browser Execution: The skill facilitates adding code interpretation and browser automation capabilities to agents. These features are designed to run in isolated microVMs or sandboxes. The instructions emphasize the use of context managers to ensure sessions are properly closed and include guidance on scoping IAM permissions to the specific resource ARNs rather than using broad permissions.
Indirect Prompt Injection Surface: The skill handles user prompts and arguments to determine which agent capabilities to configure. As it integrates tools capable of shell execution and network calls, it possesses an indirect prompt injection surface.
Ingestion points: User prompts and $ARGUMENTS in SKILL.md.
Boundary markers: The skill employs structured reference markdown files to guide agent behavior and minimize accidental obedience to embedded data.
Capability inventory: Includes shell command execution (InvokeAgentRuntimeCommand), browser automation (Playwright/Nova Act), and code interpretation (Python/JS/TS) across various reference files.
Sanitization: The process_payment_tool.py script includes SSRF validation that restricts network calls to HTTPS and blocks private/internal IP ranges. The integration guide also explicitly instructs developers to validate and sanitize inputs used in command strings.
External Resource Integration: The skill references external dependencies from standard registries and a helper repository from a known service provider for payment integrations. These resources are documented neutrally and are consistent with the skill's purpose of enabling multi-provider integrations.

agents-build