The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation (references/external-components.md) and several templates encourage the use of approximately 25 third-party GitHub repositories for extended hardware support (e.g., KinDR007/VictronMPPT-ESPHOME, syssi/esphome-jk-bms, ratgdo/esphome-ratgdo). These components are fetched and compiled at runtime by the user's ESPHome environment from unverifiable community sources.
[PROMPT_INJECTION]: The skill includes instructions in references/arduino-conversion.md for the agent to analyze and convert code from user-provided GitHub repositories. This creates a surface for indirect prompt injection, where a malicious repository could contain instructions designed to manipulate the agent's output or subvert its behavior during the analysis process.
[COMMAND_EXECUTION]: The AI agent is explicitly instructed by the 'Delivery Contract' in SKILL.md to create project folders and write multiple files to the user's disk (e.g., YAML configurations, READMEs, BOMs). This significant file-system write capability, combined with the ingestion of untrusted external code for conversion, increases the risk of the agent being used to generate or persist malicious artifacts if not carefully monitored.

ESPHome