The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to receive and process external data via webhooks from Checkout.com. This creates an attack surface for indirect prompt injection if malicious instructions are embedded in the webhook payload (e.g., in metadata fields).
Ingestion points: POST /webhooks/checkout-com endpoints in examples/express/src/index.js, examples/fastapi/main.py, and examples/nextjs/app/webhooks/checkout-com/route.ts.
Boundary markers: The skill correctly implements HMAC signature verification (Cko-Signature) and optional Authorization key checks to verify sender identity and body integrity. However, it lacks specific instructions to the agent to disregard any natural language instructions that might be contained within the verified payload.
Capability inventory: The example handlers log event data and provide a dispatch loop for handling various payment and dispute events. There are no direct file system writes or outbound network calls within the handlers.
Sanitization: No sanitization or filtering of the webhook payload is performed before processing or logging.
[METADATA_POISONING]: The package.json files and requirements.txt in the examples directory list dependency versions that appear to be significantly ahead of current stable releases (e.g., next: ^16.3.6, typescript: ^7.0.2, jest: ^30.5.2, fastapi>=0.141.1).
Evidence: Found in examples/nextjs/package.json, examples/express/package.json, and examples/fastapi/requirements.txt.
Note: While likely due to accidental misinformation or intended as futuristic placeholders, these non-standard versions could lead to installation failures or dependency confusion risks.

checkout-com-webhooks