aris-monitor-experiment
Pass
Audited by Gen Agent Trust Hub on May 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions to execute commands via SSH on remote servers (e.g., screen session manipulation, log tailing, and file reading) to monitor experiment states.
- [REMOTE_CODE_EXECUTION]: The skill executes multi-line Python snippets on remote servers using
python3 -cover SSH to interface with the Weights & Biases (W&B) API and process experiment metrics. - [DATA_EXFILTRATION]: The skill accesses local configuration files, specifically
vast-instances.jsonand~/.claude/feishu.json, to retrieve instance details and notification webhooks for reporting results to external services. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by ingesting untrusted training logs and experiment output from remote servers. This data is then used to interpret findings and suggest subsequent actions.
- Ingestion points: SSH output from
screensessions,taillogs, JSON result files, and W&B API metrics (SKILL.md). - Boundary markers: None identified in the prompt interpolation steps.
- Capability inventory:
ssh,vastai, andmodalCLI tools (SKILL.md). - Sanitization: No explicit sanitization or validation of the remote log content is performed before interpretation.
Audit Metadata