Skip to content

Run and Act on a Security Audit

This walks through a full pass through Orbit’s Security surface: run an audit, read the results, decide what to do with a finding, review and approve a fix, work a multi-finding fix plan, and set up a recurring check.

  1. Open Security from the titlebar workspace switcher.
  2. Click Audit my codebase.
  3. Select one or more registered projects.
  4. Choose a profile:
    • Quick checks — deterministic scanners against your working tree; no account needed, no model spend.
    • Deep audit — adds model-backed code review on top of the deterministic checks. Requires a ready provider account or credits; only appears enabled once one is configured.
    • History scan — runs the deterministic checks against your Git history, bounded to the 50 most recent commits.
  5. Review the confirmation screen. For a Deep audit, Orbit shows exactly what it’s about to do: redacted source, bounded to a limited file count, sent to the selected provider — read this before continuing, since it uses your account.
  6. Click Run it. You can cancel from the running screen at any point; a cancelled run is recorded as cancelled, never as a clean pass.

When the audit finishes, Security leads with the single most important open finding, then lists the rest ranked by severity, with a “smaller items” section collapsed by default. Open What Orbit checked to see the coverage breakdown — what was scanned, what didn’t apply, and where a check couldn’t run at all. If any coverage gaps exist, Security says so explicitly rather than implying a clean bill of health.

Click a finding to open its detail. From there:

  • Fix this starts a fix preview (see below).
  • Not a problem, Accept risk, or Later set the finding aside — each requires a short written reason, and “Later” needs a return date. If the finding is eligible, “Not a problem” also offers an optional checkbox to suppress future matches in that same file, project, and rule; nothing is suppressed unless you check it.
  • I’ve fixed it marks the finding as fixed pending an independent re-scan — it doesn’t close the finding on your word alone.
  • Verify with a fresh scan re-runs a targeted check against the finding’s original scope to confirm the condition is actually gone.
  • For a secret finding, Confirm rotation records that you revoked or rotated the credential with its provider — a separate step from fixing the code that exposed it.

Set-aside findings aren’t gone; find them again under Manage → Set aside, where Put back restores one to active review.

  1. From a finding, click Fix this (or Preview a fix).
  2. Security shows the exact before/after diff it proposes, scoped to one file, along with what Orbit will do automatically versus what only you can do (like rotating a credential).
  3. Click Approve this change. This is a final, explicit approval step against the specific preview you just reviewed — if the source has drifted or the preview is stale, approval fails and you’ll need a fresh preview instead.
  4. Once the change ships, use Verify with a fresh scan to confirm it actually resolved the finding.

There’s no bulk approval anywhere in Security: every fix, and every campaign step, is approved one at a time.

When an audit returns more than one priority finding, Security offers to draft a fix plan — a sequenced set of steps, secrets and shared fixes first.

  1. From the results screen, click let Orbit draft a plan.
  2. Review the plan’s phases and steps.
  3. Click Start plan to turn it into a running campaign.
  4. Work through each step: Prepare preview → review the diff → Approve this step. Rotation steps stay blocked until the code step ahead of them is verified.
  5. Verify this step once a step’s change has shipped.
  6. Use Export report at any point to get a portable Markdown copy of the plan, including any coverage gaps and rotation notes.

You can pause a running campaign and resume it later; each step keeps its own individual approval regardless.

  1. Go to Manage.
  2. Under Check automatically every week, toggle the schedule on.
  3. Choose the projects and a deterministic profile — standard or history. Security refuses to schedule a Deep audit, since a recurring model review would spend account credits without a per-run approval.
  4. Each scheduled run compares against the one before. If the schedule’s projects or scanner scope later change, the schedule pauses rather than silently comparing against an older, no-longer-matching baseline — start a fresh audit and recreate the schedule when that happens.