The Agent Skills Directory

[COMMAND_EXECUTION]: The installation flow attempts to use sudo apt-get or root privileges to install the python3-venv package if it is not present. It also modifies the user's ~/.bashrc file to append environment variables and update the PATH, ensuring changes persist across shell sessions. Additionally, it launches background processes using nohup and disown to ensure the deployed gateway remains running after the session ends.
[DATA_EXFILTRATION]: The skill reads the sensitive configuration file ~/.openclaw/openclaw.json to extract API keys and provider details. While these credentials are used to configure the local Hermes instance, accessing credential stores is a high-risk operation.
[PROMPT_INJECTION]: The soul-inject.sh script appends identity-overriding instructions to the deployed agent's SOUL.md. These instructions include a "bootstrap" section that commands the agent to perform silent self-checks and profile population on its first message. Furthermore, the skill creates an indirect prompt injection surface by interpolating untrusted data from USER.md, TOOLS.md, and memory files into the agent's context without sanitization or boundary markers.
[EXTERNAL_DOWNLOADS]: The skill fetches the uv package manager binary from the official Astral-sh GitHub repository and verifies its integrity using a SHA256 checksum.
[REMOTE_CODE_EXECUTION]: The skill clones the hermes-agent repository from the NousResearch GitHub organization and performs an editable installation using uv pip install, which executes external code within the local environment.

byreal-hermes-deploy-native