Cloud Agents

Orchestrate parallel code review and code-change runs across GitHub repositories with scoped agents, approvals, safety rules, and full run visibility. Available on Pro and above.

Cloud Agents is the repository workspace in Skytells for running specialized engineering agents against real codebases with clear scope, visible outputs, and operator control. It is designed for teams that want review work, implementation work, and repository checks happening in parallel without losing change discipline.

The product centers on three ideas:

  • Parallel work with explicit coordination. Multiple runs can operate across the same repository portfolio while Eve coordinates execution order, handoffs, and state awareness.
  • Repository-native control. Runs are attached to repositories, branches, pull requests, and approvals instead of living in an isolated chat silo.
  • Operational visibility. Every run has a status, timeline, findings view, artifacts, and approval history so teams can inspect what happened before anything lands.

Visit console.skytells.ai/agents to open the Cloud Agents workspace.

Who Cloud Agents Is For

RoleWhy this surface matters
Engineering leadsReduce queue depth, shorten time to first review, and keep change activity visible without manually coordinating every task.
Platform and developer-experience teamsStandardize how automation interacts with repositories, path rules, validation, and approvals.
Senior ICs and code ownersDelegate first-pass review and scoped implementation work while preserving final technical judgment.
Security, compliance, and release stakeholdersKeep an auditable record of what an agent inspected, prepared, or paused for approval.

Cloud Agents is most valuable when a team already has real engineering demand but wants tighter control than an ad hoc chat workflow can provide.

How Cloud Agents Works

Console prompt or GitHub signal Eve orchestration layer Review agent Execute agent Custom agents Runs Findings and artifacts Approval gates Repository outcome

Cloud Agents does not treat engineering work as a single queue. A repository can receive automated review, scoped execution, and validation workflows concurrently, with safety rules deciding what can proceed automatically and what must stop for a person.

What Teams Usually Improve

OutcomeHow Cloud Agents contributes
Faster first-pass reviewPR event signals and automated review reduce idle time before a human sees a change.
Better backlog throughputSmall and medium implementation tasks can be prepared in parallel instead of waiting for one person to cycle through them serially.
Higher change confidenceValidation-after-execute, protected paths, and approvals turn automation into a controlled process instead of a blind write path.
Clearer management visibilityRuns, approvals, and usage surfaces show where work is concentrated, blocked, or failing.

These improvements matter at both the technical and business level: teams ship more predictably, senior engineers spend less time on repetitive first-pass work, and repository owners keep the final decision rights.

When to Use Cloud Agents

PR review at scale

Use Cloud Agents when pull requests are waiting for first-pass review, style feedback, dependency checks, or scoped validation before a human reviewer steps in.

Backlog execution

Use execute runs for small and medium implementation tasks that already have clear acceptance criteria, repository scope, and rollback expectations.

Guardrailed automation

Use repository rules, protected paths, ignore paths, and approval gates when you want automation without granting unrestricted write access to sensitive code.

Operations visibility

Use the runs, approvals, and usage surfaces when engineering managers or platform teams need to understand throughput, failure modes, and quota consumption.

When Not to Use It

  • Do not start with auto-execute on a repository that has no path protection, no approval policy, and no clear ownership.
  • Do not use broad instructions such as "fix the repo" or "make everything better". Cloud Agents works best when intent, branch, and repository scope are explicit.
  • Do not duplicate existing platform setup instructions here. For GitHub authorization, billing, model access, and plan management, use the linked reference docs.

Adoption Pattern That Works

Most teams get better results when they roll Cloud Agents out in stages instead of enabling everything on day one.

  1. Start with one repository that has clear ownership and a stable pull-request workflow.
  2. Enable review-first automation before broad execute automation.
  3. Add protected paths and ignore paths before widening scope.
  4. Keep execute workflows in Plan or approval-heavy mode until the team trusts the outcomes.
  5. Expand to more repositories only after run history shows consistent quality.

Workspace Surfaces

Cloud Agents is organized around eight primary tabs in the product workspace:

TabPrimary job
OverviewHigh-level status of active and recent runs, including approval backlog and quick actions.
ChatConversational dispatch surface for scoped tasks, plans, and agent execution.
RunsFull run history with filters for queued, running, awaiting approval, completed, failed, and cancelled work.
UsageQuotas, token consumption, duration, activity charts, and repository concentration.
RepositoriesPer-repository automation defaults, safety controls, path rules, and signal settings.
ApprovalsPending human decisions for merges or other gated actions.
Add-onsCapacity extensions for review runs, execute runs, custom agents, and parallel collaboration tokens.
SettingsGlobal product setup for integrations, custom agents, and approval-related configuration.

See Workspace Tabs for the page-by-page map.

Product Model

Repositories are the control boundary

Cloud Agents does not run against a vague account-wide code universe. Automation is configured per repository. That is where you decide whether review runs can start automatically, whether execute runs are allowed, which paths are protected, and which noisy paths should be ignored.

Agents are reusable operators

Agents combine a role, a model, instructions, and operating constraints. System agents provide base review and execute behavior. Custom agents let you add repository-specific specialists for tasks such as migration work, documentation maintenance, or dependency hygiene. See Agents.

Runs are the unit of execution

Every review or execute action becomes a run. Runs preserve the input, the progress timeline, the outputs, approval state, and the final disposition. See Runs.

Eve coordinates parallelism

Cloud Agents relies on Eve to coordinate work across agents and repository state. Eve is the orchestration layer that makes concurrent execution useful instead of chaotic.

Business and Technical Fit

QuestionGood fitPoor fit
Does the team have recurring review or execution backlog?Yes, the same work patterns show up repeatedlyNo, work is highly irregular and cannot be scoped clearly
Is repository ownership clear?Yes, code owners and approvers are knownNo, nobody owns automation policy or final decisions
Is the repo operationally mature enough?CI, branch flow, and release expectations already existThe repository has no stable workflow yet
Does the team want more automation without losing control?Yes, they want speed and visibility togetherNo, they want fully unattended change generation across sensitive code

Documentation

How is this guide?

On this page