Superset is building an agent-first IDE on GitHub, wiring skills into the dev workflow
The public GitHub project under superset-sh is shipping agent skills, a Linear ticket draft flow, and editor UX tweaks, with 10.9k stars and nearly 2,900 commits.
By Ryan Merket ยท Published
Why it matters
Agent copilots are moving from sidecar chat windows into the center of the developer workflow. Superset is an example of that convergence: it treats agent capabilities as first-class IDE primitives and wires them into planning tools like Linear. If this pattern holds, the IDE itself becomes the orchestrator for human-in-the-loop automation, compressing planning, coding, and coordination into one surface.
Superset, an agent-centric IDE project maintained in the superset-sh/superset GitHub org, is pushing a steady cadence of features that make software agents a first-class part of the developer workflow.
What Superset is shipping
Recent commits suggest a clear direction: teach the editor to cooperate with agents through named, reusable "skills" and to integrate with the rest of a team's tooling. A commit titled "skill(agents): add /draft command for Linear ticket drafting" describes an in-editor flow that researches the codebase from a free-form description, drafts a ticket using a canonical format, loops with the user on accept/edit/discard, and then creates the ticket in Linear. That hints at a product that treats planning and coordination as part of the IDE, not an afterthought in a separate browser tab.
The repository includes an .agents directory where those skills live, plus agent-facing docs and formats. Separate work ties the experience to a Claude plugin surface: a refactor in .claude-plugin collapses multiple commands into a single "/superset-automation" utility for editing prompt bodies, viewing recent runs, and triggering one-off runs. Supporting changes in .claude add a shared ticket-format skill so Claude can auto-invoke it.
The team is also tending to editor ergonomics. A recent UI tweak, "Only resize on double clicking separators" (#4838), signals attention to everyday usability alongside agent plumbing.
How it works, in practice
From the commit messages and directory structure, Superset is building an IDE where:
- Agents are embedded in the editor, discoverable as skills that can orchestrate multi-step work.
- Human-in-the-loop interactions are built in via explicit ask_user steps, so agents draft and operators decide.
- External systems like Linear are accessed from within the editor flow, keeping context tight and handoffs minimal.
It is a public GitHub repository and active. The repo header shows 10.9k stars, 912 forks, 1,241 branches, 190 tags, and a history of 2,890 commits on main. The latest mainline change, commit 4462794, landed May 22, 2026.
The bigger bet
Agent features are showing up across editors and dev tools, but most offerings start as chat boxes tacked onto code. Superset is betting that agents deserve first-class primitives in the IDE itself: skills with structure, shared formats that other tools can invoke, and workflows that span reading code, drafting work, and syncing with issue trackers without leaving the editor. If that lands, it could compress the loop between what a developer intends, what the agent proposes, and what the team ships.
Open questions
There is no public founder or team profile in the materials provided. The GitHub repo is the product surface for now. Also not yet clear from the repo excerpt: licensing terms, any commercial offering, and adoption beyond stars. But the direction is legible in the commit trail, and the shipping tempo suggests the project is moving quickly toward an IDE that treats agents as collaborators, not just autocomplete.