Skip to main content
RepoFixer

Security

Trust is structural, not promised

RepoFixer's core differentiator isn't 'AI coding': it's the isolation and credential boundaries that let a coding agent work near your repository without asking your team to take it on faith.

Trust boundaries, end to end

A simplified view of how a change gets from your repository to an isolated agent and back, without ever giving the worker a direct line to your infrastructure.

  1. Your repo

    GitHub org/repo you connect

  2. Control plane

    RepoFixer's backend

  3. Worker API

    The only thing a worker can call

  4. Isolated worker

    Sandboxed execution

  5. PR back to you

    Human reviews & merges

Security architecture details

  • Execution isolation

    Agent build and deploy work runs inside gVisor-sandboxed containers. There is no fallback to unsandboxed execution. If the sandbox isn't available, the run doesn't proceed.

  • Worker authentication

    Each server RepoFixer runs agents on authenticates with its own per-server bearer token, verified on every request. That token, not a shared credential, is what the worker uses to talk to RepoFixer.

  • Data-access boundary

    Workers only ever talk to RepoFixer's Worker API. They never connect directly to MongoDB or Redis, so a compromised worker doesn't get a path to the datastore.

  • Cross-tenant credential isolation

    A worker's credentials are scoped to the jobs it's authorized to run. No worker credential is capable of reaching another customer's data.

  • Environment Vault

    Project secrets and environment variables live in a dedicated Vault, encrypted at rest, rather than being pasted into tickets or committed to source.

  • Step-up authentication

    Revealing a secret from the Vault requires a second factor at the moment of access, with a limited, time-boxed window rather than a standing, always-on reveal.

  • Customer vs. admin auth separation

    The admin console platform operators use to manage the customer fleet is a structurally separate app and auth system. Customer accounts cannot authenticate there.

  • Managed infrastructure

    RepoFixer-managed servers are provisioned under the same isolation guarantees described here: the same execution model, not a separate, looser one.

  • BYOS and its current limitation

    Bring-your-own-server keeps your own root access to the box RepoFixer provisions and runs on. Real agent execution is currently limited to RepoFixer-managed servers until BYOS reaches the same isolation guarantees. We won't claim parity it doesn't have yet.

  • Auditability

    Actions like publishing a change, connecting Jira, and revealing a Vault secret are recorded to an audit trail rather than left unlogged.

  • Human review as a control

    Every agent-produced change surfaces as a pull request. Merge authority stays with your engineers: isolation limits blast radius, but review is what actually decides what ships.

Threat-model FAQ

Found a security issue?

Email support@repofixer.com with what you found. We don't yet have a formal disclosure program or bounty, but we'll still read every report.

Contact us