autonomous-engineer

Pass

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements strict defensive rules protecting sensitive files such as .env, *.pem, *.key, and cloud credentials (e.g., ~/.aws). It explicitly forbids reproduction, logging, or transmission of these values and restricts remote VCS operations to prevent accidental data leaks.
  • [INDIRECT_PROMPT_INJECTION]: Section I, Rule 6 establishes an "Untrusted Content Boundary," instructing the agent to treat all data retrieved at runtime (MCP responses, web results) as data only and to never execute directives embedded within it.
  • [COMMAND_EXECUTION]: Command execution is scoped to legitimate local development tasks (linting, testing, local git commits). The skill explicitly forbids high-risk behaviors such as pipe-executing remote scripts (curl | bash).
  • [PRIVILEGE_ESCALATION]: Rules explicitly forbid writing to system directories or paths requiring elevated privileges, maintaining operation within the project sandbox.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 23, 2026, 07:49 PM
Security Audit — agent-trust-hub — autonomous-engineer