Skip to main content
Guardrails are real-time security checks that run inside your AI coding agent as code is being written. They stop vulnerabilities at the source, before any code reaches a pull request or gets committed.

​
How It Works

When your coding agent writes or modifies code, Asymptote intercepts the change via a hook and evaluates it against your active policy packs:
  1. The agent generates a code change
  2. The hook sends the diff to Asymptote for evaluation
  3. Asymptote scans for vulnerabilities, insecure patterns, exposed secrets, and policy violations
  4. If a violation is found, feedback is returned inline and the agent can remediate immediately
  5. The interaction is logged in Coding Sessions for audit

​
What Guardrails Catch

  • Insecure coding patterns (injection risks, unsafe APIs, hardcoded secrets)
  • Dependency additions with known CVEs or supply chain risks
  • Custom policy violations defined in natural language
  • Agent-specific failure modes like prompt injection and over-privileged tool calls

​
Setup

Guardrails are configured via the Asymptote CLI. Run asym init to connect your coding agent and apply your active policy packs. See the Quickstart for full setup instructions, or go to Config > Policy Packs in the dashboard to manage which policies are active.