Product guidance
Usage Guidelines
Last updated: July 16, 2026
Use Sentry as an additional visibility and enforcement layer around AI coding agents. These guidelines help keep a monitored session interpretable and useful.
Start with a narrow task
Describe the intended files, expected change, and any prohibited operations before starting a session. Narrow scopes create clearer predictions and make scope-drift signals meaningful.
Treat signals differently
A blocked critical action is deterministic enforcement. Scope-explosion and burn-forecast signals are monitoring signals that should prompt review or reassessment. Do not treat every warning as proof of malicious behavior.
Keep the guardrails on
Run the agent through the Foresight CLI so the Sentry proxy can observe its supported Python child process. Review the terminal dashboard or web dashboard while the session is active.
- - Use explicit allowed paths and hosts.
- - Do not disable forbidden-pattern checks to make a demo pass.
- - Stop and inspect a session before overriding a critical block.
Know the current boundary
The packaged interception path currently targets Python child processes. Other agent runtimes remain a roadmap item, so do not assume uninstrumented commands are being monitored.