Use cases.
Whatever the task, the shape is the same: a parent decides, agents do one part each on a private copy of the workspace, one reply comes back with the full trace. These are the tasks people point it at, with an honest note on how far each one is.
Repo audit
Point the parent at a repository and ask what is wrong with it. It spawns a reviewer per concern and merges the findings with path:line citations.
Test writing
A reviewer finds the behaviour worth pinning down; the test writer, which depends on it, writes the tests in the project’s own framework and runs them when allowed.
Explain a codebase
For a new contributor, a reviewer or a manager: layout, entry points, the main flow, how to run it, what looks unfinished. Always from files it read.
Changelog and release notes
From git history where the agent may run git, or from the code itself where it may not. Grouped as Added / Changed / Fixed, ready to paste.
Review a pull request
Clone the branch as the session’s workspace, review the diff by module in parallel, and post one merged review. Needs a bundle that knows your review rules.
Dependency upgrade
Read the lockfile and the changelog of the new version, find every call site that changes, write the patch and run the tests. Needs execute on the workspace.
Research brief
Split a question into sub-questions, one reader per source folder, one writer that depends on all of them. Works today on documents in the workspace; the web needs your own fetch tool.
Compare options
Vendors, libraries, designs: one agent per option reads its material, a comparer merges them into a table with the trade-offs the user asked about.
Incident triage
Drop the logs and the runbook into a workspace. A log reader, a runbook matcher and a summariser run in parallel; the reply says what happened and what to do next.
Data pipeline check
Read the pipeline code and a sample of its output, and report where the two disagree. Read-only by default, so it is safe to run on a schedule.
One task across many repos
The same task on every repository a team owns, one session each, isolated, with usage and audit per tenant. The console shows every run side by side.
Agents that spawn agents
Today one parent spawns up to four agents per message in dependency order. Deeper trees, where an agent can itself plan and spawn, are on the roadmap.
ships today the shipped bundles do it · with your bundle the parent and the runtime are ready, you write the agent · roadmap not built yet