rock-agent-debug

Pass

Audited by Gen Agent Trust Hub on Apr 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses rockcli to execute shell commands within a sandbox environment for status checks and log retrieval. While these are purposeful for debugging, they represent a significant capability tier.
  • [DATA_EXFILTRATION]: The skill accesses sandbox configuration files (.yaml, .json) and job logs. While this data is sensitive, it is processed for report generation within the local environment rather than being exfiltrated to external third-party domains.
  • [INDIRECT_PROMPT_INJECTION]: The skill reads and processes external data from sandbox files such as trajectory.json, exception.txt, and .out log files.
  • Ingestion points: Reads files from /tmp/harbor/jobs/ and /data/logs/user-defined/ (SKILL.md).
  • Boundary markers: None identified. Data is read and printed directly to the agent's context.
  • Capability inventory: The agent can execute shell commands via rockcli sandbox <id> exec and write files to the /tmp/ directory.
  • Sanitization: No escaping or validation of the log content is performed before it is output in the report.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 20, 2026, 01:21 PM
Security Audit — agent-trust-hub — rock-agent-debug