Skytells Cloud Agents
Production agents centered on GitHub repository workflows. Review pull requests, execute changes against codebases, and keep every run and output visible in the Skytells Console—triggered from GitHub and operational channels.
User-created agents - Create agents in Console. Dispatch work from a prompt box.
Cloud Agents are not a single assistant. You create agents with an advanced instruction setup (instructions + context + connected channels), then dispatch tasks from a Console-style prompt box. Each task becomes a tracked run with outputs while other agents run in parallel.
Create an agent (advanced)
Agent instructions
This is the agent’s persistent behavior guide—what to optimize for, how to format outputs, and what not to do.
Patterned after an “agent harness”: instructions + tools + model. This demo focuses on instructions and context.
Your agents
Selected agent spec (preview)
Instructions
You are a Cloud Agent focused on pull request review. Return structured findings with concrete next actions. Prefer minimal-risk changes. When unsure, ask for clarification rather than guessing.
Context
Channels
Console prompt box
Select an agent, type a task, and dispatch. The run shows up with outputs while other runs continue in parallel.
Active runs (parallel)
run_91c…12e
Review PR #128
agent repo-reviewer
run_3a8…b70
Implement webhook backoff + tests
agent feature-implementer
Each run is independent: you can inspect outputs, compare results, and keep work moving without collapsing everything into one thread.
Interactive GitHub flow - Agents operating inside real repo workflows
Click through a realistic pull-request flow to see how agents review, execute changes, and surface runs and outputs. The point is operational clarity: what happened, what changed, and where to verify it.
GitHub event
A pull request lands in the repo
Skytells Cloud Agents are centered on GitHub workflows—pull requests are a first-class trigger surface.
Fix flaky build: stabilize test runner
PR created · 3 files changed · CI failing
Diff summary
+42 −19
Review run created
Visible in Console as a tracked run with outputs and artifacts.
Highlights
- Context loaded from repository + PR diff
- Review scope set to CI failure and test reliability
Trigger
github.pull_request.opened
Mode
review
Parallel execution - Multiple agents, same repo, concurrent runs
Run more than one agent at the same time. Parallel work can target the same repository—and when your flow allows it, the same branch—while keeping each run independently visible and trackable.
Active run lane
Agent A · Bug fix
Stabilize failing CI on the active branch
- 1
Run created
Triggered from PR context
- 2
Scoped analysis
Targets the failing test + relevant diff
- 3
Patch prepared
Minimal change set; tests updated
- 4
Outcome
CI green; ready for review
Traceability
- Each lane produces an independent run record (outputs, artifacts, status).
- Results surface back into your repo flow as reviewable deltas.
Active run lane
Agent B · Feature
Implement a new repository workflow step
- 1
Run created
Triggered in Console as an implementation task
- 2
Implementation
Code + tests added against the same repo
- 3
PR output
Reviewable delta + summary artifacts
- 4
Outcome
Checks passing; awaiting review
Traceability
- Each lane produces an independent run record (outputs, artifacts, status).
- Results surface back into your repo flow as reviewable deltas.
Multi-channel command surface - Trigger runs from Slack, WhatsApp, and Telegram
Console is the operational surface, but it’s not the only entry point. Cloud Agents can receive commands from connected collaboration channels and translate them into the same run-based execution flow.
Incoming command
Commands flow into the same agent run system—so execution remains visible and trackable in Console.
Channel
Slack
Action
Create review run
Output
PR-visible findings + Console run outputs
Operational command → run
- Channel command becomes a run with explicit intent.
- Outputs are structured and visible in Console.
- Results flow back into repo artifacts (PR, patch, checklist).
What agents do - Concrete capabilities, operational outputs
Cloud Agents are designed to move work forward inside real development workflows. They’re not a generic chat UI—they act through runs and outputs you can track and review.
Review pull requests
Analyze diffs with repository context and return structured findings and next actions in the PR flow.
Execute changes against codebases
Apply focused implementation work that produces a reviewable code delta and verifiable artifacts.
Work in parallel
Multiple agents can run concurrently across the same repository—tracking each run independently.
Receive commands from channels
Trigger agent runs from connected Slack, WhatsApp, and Telegram command surfaces.
Stay GitHub-native
Operate in a PR-oriented workflow: review, patch, checks, and handoff—like real engineering work.
Maintain visibility and control
Runs, outputs, and actions remain visible so developers can validate changes and apply approvals/checkpoints per workflow.
Model choice - Powered by Skytells models and leading providers
Cloud Agents run on model configurations that match your team’s needs. Use Skytells cutting-edge models and, when desired, models from leading providers like Anthropic and OpenAI—while keeping the same run-based, GitHub-native workflow.
Skytells models
Use Skytells cutting-edge models as the default execution engine for agent runs.
Leading providers
Choose models from leading providers like Anthropic and OpenAI as part of your agent configuration.
Match model to workload
Pick the right model profile for review, implementation, or operational execution—without changing the workflow surface.
GitHub-native operation - A clear execution loop: trigger → run → output → review
Agents operate as a run system. Whether triggered from GitHub or an operational channel, work is captured as a run with outputs you can track in Console and validate in the repo.
Trigger
Run
Output
Repository event
Agent run
Visible result
PR opened, updated, or commented → run created → findings/patch shipped back to the workflow.
Channel command
Agent run
Console outputs
Slack / WhatsApp / Telegram command → run created → outputs + artifacts in Console, tied to repo context.
Docs + getting started - Start in Console, go deeper in Docs
The Console is the operational surface for creating and interacting with agents. The Docs cover configuration and usage details.
Primary
Open Cloud Agents in Console
Create and manage agent runs, inspect outputs, and keep multi-agent activity visible in one place.
Secondary
Read the Agents docs
Learn configuration details, workflow triggers, and how to use agent runs effectively across repository and channel entry points.
FAQ - Answers developers actually look for
Clear, operational answers about how Cloud Agents work in GitHub workflows, how runs are tracked, and how model choice fits into execution.
What are Skytells Cloud Agents?
+
What are Skytells Cloud Agents?
+Skytells Cloud Agents are production agents centered on GitHub repository workflows. They operate through tracked runs and outputs—reviewing pull requests, executing changes against codebases, and surfacing results in the Skytells Console.
Are Cloud Agents just a chat assistant?
+
Are Cloud Agents just a chat assistant?
+No. Cloud Agents are designed for repository workflows and operational execution. Work is represented as runs with outputs and artifacts you can inspect, review, and keep moving through PR-oriented flow.
Can I create my own agents?
+
Can I create my own agents?
+Yes. You can create agents in Console with an advanced setup (instructions, context, and connected channels), then dispatch tasks from a prompt box. Multiple agents can run in parallel.
Which models power Cloud Agents?
+
Which models power Cloud Agents?
+Cloud Agents can be powered by Skytells cutting-edge models and models from leading providers like Anthropic and OpenAI, depending on your configuration.
How do agents work with GitHub pull requests?
+
How do agents work with GitHub pull requests?
+A repository event or command creates an agent run. The agent reviews diffs with repository context, produces structured outputs (findings, checklists, patches), and the result is visible in Console and in PR-oriented flow.
Can agents receive commands from Slack, WhatsApp, or Telegram?
+
Can agents receive commands from Slack, WhatsApp, or Telegram?
+Yes. Commands can be received from connected operational channels (Slack, WhatsApp, Telegram) and translated into the same run-based execution flow so outputs remain visible and trackable in Console.
Put agents to work in your GitHub workflows
Start in Console. Keep runs visible. Review code deltas like a normal PR.