The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/server_access.py uses subprocess.run to execute the ssh binary. It provides an inspect subcommand that allows the agent to run arbitrary shell commands on remote servers.
[DATA_EXFILTRATION]: The skill accesses sensitive local files, including the SSH configuration and identity keys at ~/.ssh/id_ed25519 and a central credentials file at ~/.systemcraft/ssh/creds.yml. The script outputs the path to the identity key in its standard response, which reveals details of the local security environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes output from remote servers without sanitization or clear boundary markers (ingestion points: scripts/server_access.py). This remote output is fed directly into the agent's context, potentially allowing a compromised server to influence agent behavior. The skill has significant local capabilities via the ssh command (capability inventory: subprocess.run in scripts/server_access.py, sanitization: none for remote output).

server-access