nemo-rl-brev-etiquette
Pass
Audited by Gen Agent Trust Hub on Jun 12, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill instructs the agent to load sensitive credentials (such as WANDB_API_KEY and HF_TOKEN) from a local environment file at
/home/ubuntu/RL/.env. This is a standard and recommended practice for secret management. - [DATA_EXFILTRATION]: The instructions include a specific security constraint: 'Never print, cat, log, commit, or summarize secret values', which explicitly directs the agent to avoid common exfiltration or exposure patterns.
- [COMMAND_EXECUTION]: The skill provides bash code blocks for checking disk space (
df -h), creating directory structures (mkdir -p), and exporting environment variables. These operations are within the scope of the skill's intended purpose for environment setup. - [EXTERNAL_DOWNLOADS]: The skill-card references official NVIDIA documentation and GitHub repositories. These are verified vendor resources and do not involve untrusted third-party downloads.
- [PROMPT_INJECTION]: The skill contains instructional content but no patterns associated with bypassing safety filters, ignoring instructions, or extracting system prompts.
Audit Metadata