The Agent Skills Directory

Command Execution (Inter-process Bridge): The skill utilizes a bridge between its TypeScript and Python components. In packages/openclaw/src/bridge.ts, child_process.spawn is used to execute a specific Python script (agentcore_bridge.py) for AWS service calls. This is implemented without a shell and uses fixed paths to minimize risk.
Command Execution (Test Infrastructure): Several test scripts, such as scripts/test_x402_policy.py and scripts/test_portability.py, utilize subprocess.run to verify the functionality of the CLI and ensure skill portability across different agent harnesses.
External Package Dependencies: The skill requires several standard Python packages listed in requirements.txt, including boto3 for AWS services and httpx for networking. These are well-known libraries necessary for the skill's primary functionality.
Data Ingestion and Isolation: The skill processes potentially untrusted content from paid HTTP responses. To mitigate indirect prompt injection risks, it withholds response bodies by default. If enabled by an operator, bodies are truncated to 10 KiB and explicitly marked as untrusted: true in the agent's context.
SSRF Protection: The payment retrieval tools in scripts/x402_fetch.py and packages/openclaw/src/x402.ts implement comprehensive protections against Server-Side Request Forgery. This includes resolving and pinning IP addresses to prevent DNS rebinding attacks and blocking access to private, loopback, and reserved IP ranges.

agents-pay