TRMNL

The SKILL.md states the skill is for sending content to a user's TRMNL e-ink device, which is coherent with using a webhook. However the webhook URL is hardcoded in the instructions (https://trmnl.com/api/custom_plugins/0d9e7125-789d-46a6-9a51-070ac95364d8) and there is no mechanism (no required env var, no config path, no install-time prompt) for the installing user to supply their own webhook or authenticate. That means messages from any user installing this skill would be posted to the single embedded endpoint rather than to a per-user device — a mismatch between claimed purpose and actual configuration.

The runtime instructions tell the agent to POST arbitrary user-provided content to the fixed external URL via curl. There is no guidance to confirm device ownership, no sanitization or filtering, and no restriction against sending secrets. This grants the skill the ability to transmit any text the user (or agent) supplies to a third-party endpoint.

No install spec and no code files: the skill is instruction-only and does not write files or download code. This is the lowest install risk.

The skill declares no environment variables or credentials, which would normally be expected if the skill needed to target a user's personal webhook or authenticate to a device. The absence of configurable credentials combined with a hardcoded webhook is disproportionate and suspicious.

The skill is not always-enabled and requests no persistent system privileges. It does not modify other skills' configs or system settings based on the provided metadata.