remote-mac
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from various local files and tool outputs which could theoretically contain instructions intended to influence the agent's behavior.
- Ingestion points: The skill reads from
/Users/steipete/Projects/manager/computers.yaml,/Users/steipete/Projects/manager/agents.yaml, and processes JSON output fromtailscale statusandopenclawcommands. - Boundary markers: No explicit delimiters or instructions are provided to the agent to treat data from these sources as untrusted or to ignore embedded instructions.
- Capability inventory: The skill is designed to execute remote shell commands using
sshand read local files. - Sanitization: No sanitization or validation of the content of the inventory files or tool outputs is specified before the data is used to inform agent actions.
- [DATA_EXFILTRATION]: The skill directs the agent to access sensitive infrastructure metadata which constitutes a data exposure risk.
- Evidence: The instructions involve accessing
computers.yamlanddocs/tailnet-portal.md, which contain hardware UUIDs, private node IDs, and network topology details. While it includes instructions not to publish this data, the agent's ability to read it creates an exposure surface. - [COMMAND_EXECUTION]: The core functionality of the skill involves executing commands on remote hosts.
- Evidence: The skill provides specific patterns for executing non-interactive
sshcommands and running scripts on multiple remote Mac Studio and Mac mini devices.
Audit Metadata