config-hardener

Pass

Audited by Gen Agent Trust Hub on Mar 11, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill disables network and shell permissions, preventing external data exfiltration and unauthorized command execution.
  • [SAFE]: The instructions explicitly promote security hardening and prohibit the agent from accessing sensitive credentials or system files.
  • [PROMPT_INJECTION]: The skill processes untrusted local data, creating a potential surface for indirect prompt injection. * Ingestion points: Reads local configuration files such as AGENTS.md using the fileRead permission. * Boundary markers: There are no explicit delimiters or markers to isolate file content from instructions. * Capability inventory: The skill has fileRead and fileWrite permissions to perform its audit and hardening tasks. * Sanitization: The skill does not perform validation or sanitization on ingested file content before it is processed by the model.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 11, 2026, 01:19 AM