Put your AI agents
on a harness.
Every agent runs the same workflow: research, plan, implement, validate, review. You approve the spec before an agent writes code, and you read the diff before it merges.
One harness, end to end
You control the spec an agent follows, the worktree it runs in, and the gate it clears before merge.
Specs your agents follow
You run /research_codebase and /create_plan, then read the plan they draft: phased work, explicit success criteria. Sign it off before an agent writes a line of code. The plan lives in a git-backed thoughts repo your teammates can read.
- ✓One version-controlled skill set your team shares
- ✓Plans live in a shared, git-backed thoughts repo
- ✓Hyprlayer picks Opus for research and planning
The whole fleet on one screen
Launch and steer your agents from one window. Each one gets its own git worktree and branch, so two teammates can ship in parallel without touching each other's files. The CLI and the desktop app run the same engine.
- ✓Isolated git worktree + branch per agent
- ✓Live +/− diff counts on every workstream
- ✓Drives Claude Code today; Codex coming soon
Agents clear the same gates you do
You run /validate_plan to check the finished work against the criteria you signed off on, then /code_review to open a PR and track CI. Read the diff and merge it the way you would a teammate's.
- ✓Checks the work against the plan's success criteria
- ✓Diff, staging, and one-click PR with CI status
- ✓Same review gates as human-written code
Terminal-native or desktop cockpit
The same skills and the same spec-driven workflow, for Claude Code, Copilot, and OpenCode. You stay in the terminal.
Boards, diffs, and live agent streams in one native window.