The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill mentions using well-known services such as ngrok and Cloudflare Tunnel for creating public HTTPS endpoints during development. These are established tools for webhook testing and do not pose a security risk in this context.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch public signing keys from official J.P. Morgan infrastructure at mns-aws.jpmchase.com and mns-aws-cat.jpmchase.com. These are legitimate vendor domains used for cryptographic verification of incoming data.
[INDIRECT_PROMPT_INJECTION]: The skill handles incoming webhook data, which is an external data ingestion surface. However, it provides extensive security instructions to mitigate risks.
Ingestion points: Raw request bytes and headers (Signature, Key-ID, Signing-Algorithm) from inbound POST callbacks.
Boundary markers: Explicit instructions to verify signatures before any parsing or business logic occurs, ensuring only authenticated payloads are processed.
Capability inventory: The skill does not execute subprocesses, perform dynamic code evaluation, or engage in unauthorized network operations. File operations are restricted to project-specific paths.
Sanitization: Implementation of SHA256withECDSA or RSA-SHA256 signature verification ensures the integrity and authenticity of the received data, effectively preventing forged instruction injection.

jpm-notifications