Skip to content
NewSee how fast an edit goes live on this site
Halftone

SecurityEngineering

Our approach to sandbox isolation, part 3

How we make the safe path the easy path.

Lu Brandt / Kit Moreau
Published 2 min read

We publish what we change and why. Customers should be able to check our work, not just take our word for it. This post covers what we changed, why, and what it means for you.

#The constraint

We assume something will go wrong and design for recovery. Containment, rollback and clear alerts matter more than any single control.

Every action is attributed to a person or an agent acting for one. The audit log answers who did what, when, and with which permissions.

Credentials never reach the code that runs on your behalf. A broker holds them and hands out short-lived tokens scoped to a single task.

#What changed

Permissions only shrink during a run. A task can give up access it no longer needs, but it can never gain access it did not start with.

Security work is mostly about making the safe path the easy path. If doing the right thing takes extra steps, people will eventually skip them.

We publish what we change and why. Customers should be able to check our work, not just take our word for it.

ts
const key = `${route}:${hash(content)}:${hash(template)}`;
const hit = await cache.match(key);
if (hit) return hit;

#Results

We assume something will go wrong and design for recovery. Containment, rollback and clear alerts matter more than any single control.

Every action is attributed to a person or an agent acting for one. The audit log answers who did what, when, and with which permissions.

#Trade-offs

Credentials never reach the code that runs on your behalf. A broker holds them and hands out short-lived tokens scoped to a single task.

Permissions only shrink during a run. A task can give up access it no longer needs, but it can never gain access it did not start with.

Security work is mostly about making the safe path the easy path. If doing the right thing takes extra steps, people will eventually skip them.

  • Cache what never changes
  • Run independent steps in parallel
  • Keep the previous version one click away

#What we would do differently

We publish what we change and why. Customers should be able to check our work, not just take our word for it.

We assume something will go wrong and design for recovery. Containment, rollback and clear alerts matter more than any single control.

Diagram of the request path before and after the change
Figure: the request path, before and after.

#Where we started

Every action is attributed to a person or an agent acting for one. The audit log answers who did what, when, and with which permissions.

Credentials never reach the code that runs on your behalf. A broker holds them and hands out short-lived tokens scoped to a single task.

Permissions only shrink during a run. A task can give up access it no longer needs, but it can never gain access it did not start with.

  • Cache what never changes
  • Keep the previous version one click away
  • Skip what has not changed