soia-dev-terminal-ops
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill enables management and execution of background shell commands via tmux. It includes specific safety guidelines to prevent command injection, instructing agents to avoid direct shell fragment splicing and to use host-tool safe transmission for user-provided inputs.
- [SAFE]: The skill implements a comprehensive diagnostic framework to assess process health using log activity, CPU usage, and network state before recommending termination. It uses a tiered confirmation gate (TERM signal, verification, then KILL signal) to prevent data loss or accidental process interruption.
- [DATA_EXFILTRATION]: No exfiltration behaviors were detected. The skill uses standard diagnostic tools like
psandlsoffor local state monitoring and explicitly warns against logging or displaying sensitive information such as authentication tokens or secrets. - [EXTERNAL_DOWNLOADS]: Installation instructions and documentation references target the author's verified GitHub repository and package namespaces (soia-team), representing standard vendor functionality for skill deployment.
- [SAFE]: Regarding indirect prompt injection risks:
- Ingestion points: The skill reads external data from terminal logs (
tail), tmux pane history (capture-pane), and process network metadata (lsof). - Boundary markers: Instructions define a structured markdown response format to separate diagnostic signals from command outputs.
- Capability inventory: Capabilities are limited to process management (
kill), session control (tmux), and diagnostic logging. - Sanitization: The instructions mandate desensitization of sensitive parameters and warn against executing unverified shell fragments.
Audit Metadata