The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the remote_bash tool, which allows the agent to execute arbitrary shell commands on remote cluster nodes. This is implemented in mcp-server/mcp_remote_server.py and cluster-agent/agent.py using Python's subprocess module to interface with SSH and bash.
[REMOTE_CODE_EXECUTION]: The skill automates the deployment and execution of code on remote systems:
In SKILL.md (Step F), the agent is instructed to deploy the agent.py script to a remote cluster by writing its contents through a shell heredoc.
The mutagen-setup.sh script extracts the Mutagen agent binary from the local environment and uploads it to the remote cluster for execution over SSH.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes data from external systems.
Ingestion points: Data enters the agent's context through the stdout and stderr output captured by the remote_bash tool in mcp-server/mcp_remote_server.py.
Boundary markers: The skill does not implement delimiters or specific instructions to the agent to disregard potential commands embedded within the remote output.
Capability inventory: The agent possesses extensive capabilities, including remote filesystem access, remote execution via remote_bash, and local MCP server registration via setup.sh.
Sanitization: While outputs are decoded with character replacement for errors, there is no semantic sanitization or filtering of the content returned from the remote cluster.
[COMMAND_EXECUTION]: The mcp-server/setup.sh script performs local system modifications by executing claude mcp add or codex mcp add to register the MCP server, injecting JSON-encoded environment variables into the agent's configuration.

remote-cluster-agent