Agentic is pre-1.0 software. Security fixes are applied to the latest commit on main.

Please use GitHub's private vulnerability reporting for this repository. Do not open a public issue for leaked credentials, unsafe automation behavior, or a permission bypass.

Include:

• the affected commit
• reproduction steps
• the expected and observed behavior
• the potential impact
• a suggested mitigation, if known

Agentic can control local applications and browsers. Treat every skill, configuration change, and automation adapter as trusted code.

• Keep .env files and access tokens out of Git.
• Review actions before execution.
• Use the emergency stop shortcut: Ctrl + Alt + Q.
• Do not run unreviewed community skills.
• Do not expose the local API to the public internet.
• Keep external publishing and destructive actions behind explicit permission.

Screen text, browser pages, OCR, and extracted content are untrusted inputs. Agentic labels them accordingly and validates every resulting action, but prompt injection defenses are not perfect. Keep allowlists enabled and do not use the pre-1.0 desktop agent for financial, administrative, or other high-impact tasks.

Privacy Mode masks the Agentic interface during a screen share. It does not bypass recording software or operating-system monitoring.