What Ona Agent can do
- Write and modify code: features, bug fixes, refactors, tests, documentation
- Run your toolchain: commands, formatters, linters, test suites in isolated environments
- Manage pull requests: open branches, create PRs, respond to review feedback
- Execute long-running work: migrations, large refactors, multi-repo changes (see Automations)
- Follow your policies: command allow/deny lists, SSO, audit logging
How Ona Agent works
Give Ona Agent a task through chat, a Linear issue, or an automation trigger. It spins up an isolated environment using your Dev Container configuration, works through the task on its own, and opens a pull request when it’s done. Each environment is isolated from your machine and other environments. If something goes wrong, discard it and start fresh. For how agents fit into the broader platform, see Core Components.Prerequisites
Configure Ona Agent
- Teach agents your codebase: create an AGENTS.md with your conventions, or add Skills for repo-specific workflows
- Organization-level skills: codify your team’s workflows with skills that agents discover proactively
- Connect integrations: enable Linear or other integrations for richer context
- Set guardrails: configure command deny lists, executable deny lists, and policies for safe execution
Best practices
- Start with smaller tasks and expand scope as you build confidence
- Keep environments reproducible with Dev Containers and Tasks
- Use guardrails for safe, auditable execution
- Test workflows in non-production repositories before broad rollout
Next steps
- Codex Agent: move workloads to the open-source Codex harness
- AGENTS.md: teach agents your conventions
- Skills: create reusable prompts agents discover proactively
- MCP servers: extend agent capabilities
- Command deny list: restrict dangerous commands
- Executable deny list: block specific binaries at the kernel level