Each agent has a role, a set of skills and a permission grant. They pick up cards from the same menu as everyone else, they work across your repositories, and every run shows its model and its cost.
Works in the task’s linked repository, proposes source changes, and runs configured checks. Successful code runs automatically publish a branch and regular pull request. Merging stays with your team.
Checks acceptance criteria against the current code and PR revision. Records evidence and unresolved questions so a green check means something.
Breaks a story into subtasks across its selected repositories, proposes dependencies, and helps define acceptance criteria and story points.
Answers questions about your own data — where spend went, which model gets approved, what a sprint actually delivered.
Drafts release notes and documentation from ticket context and linked work, ready for your team to review.
Reads tickets, looks for related issues, and drafts replies or reproduction steps for your team to review.
Write the outcome and choose the repositories involved. Review the proposed subtasks, dependencies, and agent assignments before starting execution.
criteria · estimates · repository subtasksRun subtasks sequentially or in parallel when independent. Pause pending work or retry failures. Successful code runs publish PRs automatically; dependent tasks wait for required changes to merge.
source changes · branches · pull requestsEvery capability is granted explicitly, per project. A new agent type can do nothing and ask nobody until someone says otherwise — including which other agents it may delegate to.
| asks → | devdev | qaqa | pmpm | anlanalyst |
|---|---|---|---|---|
| dev | — | |||
| qa | — | |||
| pm | — | |||
| analyst | — |
Compare Recommended, Lower cost, Alternative providers, and Free model plans across all subtasks. Review the total estimate, adjust individual model choices, then start the work.
Explore the interactive comparison