AI CodeOps

Weekly codebase inspection on a schedule

Schedule Swarm Inspection to sweep one module every week: reviewer, security and tester slices merged into one summary, plus a fix for the worst issue it finds.

Use this template

Code review sees what changed. It doesn't see the module nobody has touched in a year. This template sweeps one area of the codebase on a schedule and turns the worst problem into a change you can review.

Which agents run

StepAgent
Reviewer sliceReviewer
Security sliceSecurity Auditor
Tester sliceTester
Merge summaryArchitect

Each slice looks at the module from one angle: correctness, security risk (authorization, injection, data exposure, supply chain) and missing tests. The Architect merges them into one ranked summary, which you read on the session page.

The code pass then acts on the task: here, a fix for the most serious item. Keep the module small enough to inspect in one session; agents read a selected snapshot of the repository, not the whole tree.

Gates that apply

The fix goes through the usual quality gates: your test command runs against it, the findings review reads its diff with the slices as context, open high or critical findings block, and a person approves it. Use Fix on a finding to start a session that fixes it.

The rest of the summary is advice. To act on more items, send a follow-up from the session page, one item at a time.

Schedule it

Open Release → Schedules and choose New schedule:

FieldExample value
NameWeekly payments inspection
WorkflowSwarm Inspection
Projectpayments-api
Cron expression0 6 * * 1 (Mondays at 06:00)
TimezoneEurope/London
Taskthe task above
If a tick is missedSkip missed (fire only on schedule going forward)

Choose Preview next 5 to check the times, then Create schedule. Run now on the row starts one immediately, and History lists past runs. Rotate the module in the task every few weeks so the sweep covers the whole codebase.

Every run costs model spend, so a per-workflow cap in cost controls keeps a forgotten schedule in check.

Frequently asked questions

Do the swarm slices run concurrently?

No. Swarm describes how the work is split into slices. The slices run one after another, and the merge step reads them all.

Does every scheduled run open a pull request?

No. Each run proposes a change and opens an approval request. Nothing becomes a pull request until a person approves it and the gate passes.