rollback

Pass

Audited by Gen Agent Trust Hub on Jul 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes dynamic context injection (!kubectl config current-context) to execute shell commands at load time. This retrieves and exposes the active Kubernetes context to the agent, which is used to verify the deployment environment.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and analyzes untrusted data from external sources.
  • Ingestion points: The skill reads deployment targets, PR references, and commit messages (found in SKILL.md under 'Input' and 'Behavior').
  • Boundary markers: No specific boundary markers or 'ignore' instructions are used to separate the external data from the agent's instructions.
  • Capability inventory: The agent can write findings to the local filesystem using the obsidian_append_content tool.
  • Sanitization: There is no explicit sanitization or validation of the PR/commit content before it is processed and documented.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 26, 2026, 12:10 AM
Security Audit — agent-trust-hub — rollback