The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and installs executable skill definitions into the agent's environment.
It clones remote content and copies it directly into the agent's active skill directories (~/.openclaw/skills/ or ~/.hermes/skills/).
These installed files are subsequently loaded and executed by the agent in future sessions.
[COMMAND_EXECUTION]: The skill uses shell commands for repository and file system management.
It executes git commands (clone, fetch, log, describe) to manage version control and synchronization of the skill bundle.
It performs directory operations including moving, copying, and deleting files within the user's home directory.
[EXTERNAL_DOWNLOADS]: Fetches updates from a remote Git repository.
The default source is the vendor's repository at github.com/OpenSenseNova/SenseNova-Skills.git.
The skill instructions explicitly allow for user-provided URL overrides (forks), which introduces a potential vector for installing untrusted code.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a gateway for external content to influence agent behavior.
Ingestion points: External Markdown and script files from the remote repository.
Boundary markers: None identified; the skill content is integrated directly into the agent's execution path.
Capability inventory: Network access (git), file system write/delete access, and the ability to modify the agent's own logic (self-updating).
Sanitization: There is no evidence of integrity checking (e.g., GPG signatures) or content sanitization for the downloaded files beyond Git SHA comparisons.
[PERSISTENCE]: Modifies the agent's permanent installation to maintain functionality across restarts.
By installing and updating files in the agent's core skill directories, the skill achieves persistence for the new or modified code.

sn-update