The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to "work around the security guard" designed to prevent direct editing of sensitive configuration files. This directs the AI to circumvent platform-enforced safety and integrity constraints.
[REMOTE_CODE_EXECUTION]: The skill includes a pattern that pipes data from a network endpoint directly into the Python interpreter: curl -s http://127.0.0.1:8642/api/model/info | python3. This is a critical risk vector as it executes unverified content from a network source.
[COMMAND_EXECUTION]: The instructions rely on the use of shell here-documents to execute dynamically constructed Python code (e.g., `python3
<<'PY' ... PY`). This dynamic execution of scripts at runtime increases the risk of command injection and malicious logic execution.
[DATA_EXFILTRATION]: The skill targets sensitive configuration files such as ~/.hermes/config.yaml and .env which contain authentication tokens and API keys. It provides specific instructions for reading and displaying these values, creating a risk of credential exposure.
[EXTERNAL_DOWNLOADS]: The skill references an external patch script (apply-patches.sh) to modify the agent's core source code and uses network requests to fetch configuration metadata, which introduces unverified external logic into the local environment.

hermes-config-editing