Use casesCoding agent controls

Solution brief 01 · Governance · Security

Policy Enforcement for Coding-Agent Actions

For Engineering, AppSec, Platform security

The problem

  • Coding agents run real shell commands with a developer’s access.
  • The same operation can be worded many ways, so command lists miss it.

What MACAW does

  • Denies rm -rf under an admin policy, however the request was worded.
  • Holds a curl call until a person approves it, then runs it on the signed approval.
  • Puts both decisions on the audit timeline.

What’s different

  • Policy is about what an operation does, not how it’s spelled.
  • Allow, deny or wait for approval, per operation.
  • Enforcement and approval sit outside the agent’s reasoning.

Policy and deployment

Policy
rm -rf is never allowed. Network access waits for a person.
Deploy
The MACAW plugin and sidecar beside Claude Code or Codex. npm install -g macaw-ai, then macaw-ai install.

MACAW · AI Controls for autonomous agents. Every action an agent takes, checked against policy before it runs. Every decision signed.