AI CodeOps

Agents and workflows

The AI agents that run in an AI CodeOps session and the 16 built-in workflows that order them, from Plan, Build, Review & Test to Incident Response.

A session isn't one AI model answering one prompt. It is a workflow: a fixed sequence of steps, each run by an agent with its own role and instructions. Planning, building, reviewing and testing are done by different agents, so the work is checked by something other than the thing that wrote it.

The built-in agents

AgentRole
ArchitectPlans, risks, staged execution.
BuilderImplements approved changes.
ReviewerCorrectness and maintainability review.
TesterTests, typecheck and quality gates.
Security AuditorSecurity risks, secrets, CVEs.
UI DesignerVisual and UX planning.
DevOps EngineerCI/CD, infrastructure, pipelines.
Product AnalystRequirements and acceptance criteria.
Project PlannerPhased delivery plans.
ResearcherPrior art and codebase survey.
Backend EngineerServer-side implementation.
Frontend EngineerClient-side implementation.
E2E TesterEnd-to-end test coverage.
Dependency AuditorSupply-chain audit.
Database EngineerSchema and migration design.
Incident ResponderIncident triage.
Site ReliabilitySLOs, runbooks, postmortems.
ADR WriterArchitecture decision records.

Some workflows also use a technical writer, a migration specialist and a performance tester. More about each agent is on the agents page.

The built-in workflows

Pick a workflow in the Workflow list when you start a session. Develop → Workflows shows the full catalog.

WorkflowSteps
Plan, Build, Review & Test (default)Plan → Build → Review → Test
Quality-Gated ChangeBuild → Test → Review
Security-Sensitive ChangeScope change → Risk analysis → Implement → Audit diff → Validate
Parallel ReviewCorrectness, security and test-coverage reviews, one after another, then findings merged
UI ImplementationUI plan → Implement → Code review → Behavior test
DevOps ChangeDevOps plan → Risk review → Apply → Validate
Feature SDLCRequirements → Delivery plan → Technical design → Backend → Frontend → E2E tests → Security review → User docs
Discovery & SpecPrior art → Requirements → Delivery plan (no code)
Architecture DebateTwo architects propose, a reviewer critiques, the architect synthesises
Threat ModelScope & data-flow → STRIDE analysis → Record ADR
Swarm InspectionReview, security and test slices across the codebase, then a merged summary
Release ReadinessReview, security, E2E sanity and dependency audit, then a go/no-go
Incident ResponseTriage → Root cause → Mitigation → Verify → Postmortem
Dependency RefreshAudit → Upgrade → Test → Verify secure
Performance TuningProfile → Optimization plan → Apply → Re-measure
Data MigrationSchema → Migration → Validate → Security → Final review

Architecture Debate, Discovery & Spec and Threat Model record an architecture decision record (ADR) you can find under Plan → ADRs.

When the code is written and checked

The code is written after the step that implements the change (Build, Implement, Apply, Mitigation, Upgrade, Migration, or Backend and Frontend), and your project's tests run on it. The steps after that one, such as Review, Test, Audit diff or Re-measure, run next. They are shown the change as written and the recorded test result, so they check the code, not a plan. A workflow with no implementing step, such as Parallel Review, runs all its steps before the code is written.

Whatever the workflow, a session that writes code goes through the same checks: your project's tests run against the change, an AI review records findings, and a person approves it before a pull request opens. See Quality gates.

Models per step

On the platform's models with Auto, each step gets the tier that suits it: steering steps (architect, reviewer, security auditor, researcher, planners) get the best tier you have allowance for, code-writing gets Standard, and drafting gets Economy. See Models and tiers.

Custom agents and skills

Under Configure → Agents, choose New agent to define your own agent: its model, permissions and prompt.

A skill is a set of reusable instructions that a session's agents follow when it is relevant. Sessions load two kinds:

  • Repository skills. A SKILL.md file in a folder under .codeopsstudio/skills/ (or .claude/skills/) in a project's repository. Add or change one by committing it. Every session on that project loads it.
  • Built-in skills. Documentation, Security Review, Refactor, Migration and Test Generation. They are off until you turn them on: under Configure → Skills, pick a project and choose Turn on next to a skill. From then on, every session on that project loads it. Turn off removes it again. You need permission to manage projects to change this, and each change is recorded in the audit log.

If a project's repository has its own skill with the same name as a built-in one, sessions use the repository's. A session started without a project loads no skills.

Policies can limit which agents and workflows run without approval. See Approvals and policies.

Next steps

Frequently asked questions

Which workflow should I start with?

Plan, Build, Review & Test. It is the default and suits most code changes. Pick a specialised workflow when the task is clearly security, UI, infrastructure, data or incident work.

Can I add my own agents?

Yes, for the built-in roles. Under Configure → Agents you can give an agent your own prompt, model and permissions; it takes effect where a workflow step uses that agent. Workflows themselves can't be customised yet, so an agent for a brand-new role isn't run.

Was this page helpful?