Paireto brings pair-programming to your TUI coding agents in VS Code: planning, review, editing, and worktree management, in one engineer-grade workflow. Contents
Why I Built PairetoAgentic coding changed software over-night. But for high-craft engineering, the direction between tools and the reality of engineering work keeps widening. I’m an engineer working with engineers. My code still has to meet a human standard. I don’t need a vibe-coding tool. I tried the Conductor immitators, with worktrees, git diffs, and agent sessions alongside my editor. They were useful, but high-friction. The moment I needed LSP feedback, linter errors, or a quick manual edit, I was back in VS Code. They were a second app with weaker versions of features my editor already had. I tried GUI agents, but they pulled me toward more mouse-driven workflows or a new editor ecosystem. I tried TUI agents, and loved the ease they integrated with my editor workflows, but they struggle with structured planning, review, editor integration, and human reading comprehension. I increasingly found myself editing code from Git Diff tabs, so I could track what the agent had changed and refine. But had no way to prepare agent feedback easily. I tried Plannotator, and it clicked: planning and review are two missing pieces in my workflow. But I still had to jump between apps, and work outside my editor during code review, losing LSP features which I depend on to understand code. I didn't need a second app, I needed a tighter integration between my TUI and editor. So I built this... Features at a glance
InstallationPaireto comes in two parts:
Agent harness setupOn first install a Welcome wizard will take you through setup of your agents. You can return to
this screen at any time from the Command Palette by opening After plugin setup, restart your agent to load the Paireto integration. Open a repo in VS Code, start the agent in its terminal, and the agent appears in the Paireto sidebar. Agent supportPaireto's architecture is agent-agnostic, but still in development. We currently support
WorkflowsPlan ModeWhen your agent finishes planning, the plan opens in VS Code and the agent waits. You may leave inline comments from the VS Code editor, then click Approve or Send Feedback to instruct the agent Review ModeWhen your agent tries to end its turn with any changes made, a review is (by default) started automatically. You can also start a review at any time with the Kiro is the exception: it has no automatic review. Kiro runs its Stop hooks once per agent run,
so the one pass is usually spent before the work is finished and a review gated on it opens late or
not at all. With Kiro, ask for the review yourself — Diffs are fully functional editors with LSPs and linters working as normal. Add inline comment from VS Code edit or diff tabs. Click Send Feedback to hand over your comments, or Approve to let the agent finish. Guided ReviewAsk your agent for a guided review ( A Review Plan section replaces the Changed Files list while it is open. Work through it changeset by changeset: click a changeset to read a description of what it does, click a file to open its diff, and stage or unstage a whole changeset from its row. Each file row shows which git layer it sits in (committed, staged, or working tree). Anything the plan did not name is collected under Other changes, so nothing is hidden. Comment on a file diff as usual, or on a changeset's description to give feedback on the grouping itself. Finish with the usual Approve or Send Feedback. Changed Files viewChanged Files is the native git panel with extras: pick what to Compare To (HEAD, merge-base, stack-base, default branch, or any other ref), add review comments for your agent, and step in manually to fix code. Switch repo or worktreeHit Troubleshooting
|
