Use cases
The same platform, three jobs. Ship changes from an issue to a pull request, review every push before it merges, and put the recurring work on a schedule.
One command from the repository you are standing in. The agent gets its own sandbox, runs your suite, and opens the pull request.
$ agent session start "add rest-period warnings to the schedule editor" --watch
using config "Feature builder" (repo default)
session_7Hq2mX4p · sandbox ready in 8s · splitshift-web cloned
● Read src/schedule/editor.tsx
● Grep "restPeriod" (14 matches)
● Edit src/schedule/rest-period.ts
● Edit src/schedule/editor.tsx
● Bash (npm test)
⎿ Tests: 214 passed, 214 total
✓ opened PR #514 · 6 files · +212 −18
session_7Hq2mX4p · 9 turns · 88.4k tokens · $0.94
The prompt, the model, and the repositories it watches live in one file you own.
ellipsis: version: v1 kind: code_review name: Code reviewpull_requests: repositories: [splitshift-web]review: - name: correctness claude: model: claude-opus-4-8 system: | Flag correctness bugs and unhandled edge cases. Fix on the exact line.budget: run: 10.00Describe the rule and pick when it fires. Path filters mean the agent only wakes up on the pull requests it applies to.
“Make sure every pull request that changes the frontend has a screenshot of the change.”
● Bash (npx playwright screenshot)
⎿ schedule-editor.png · 1440×900
✓ commented on #518 · $1.42
ellipsis: version: v1 name: Screenshot checkclaude: model: claude-haiku-4-5-20251001 system: | Start the dev server, open the affected page with playwright, and comment with the screenshot.trigger: type: react pull_request: on: [opened, pushed] repositories: [splitshift-web] paths: ["src/**/*.tsx"]sandbox: repositories: - name: splitshift-web image: setup: | cd splitshift-web && npm install npx playwright install --with-deps chromium github: permissions: contents: read pull_requests: writebudget: session: 3.00Define in YAML, deploy with git push
Starter YAML templates for common tasks
You can close your laptop now
An isolated computer for every session
Log every step, including Thinking and Tool calls
Define agent permissions as code
Engineering metrics for the age of AI
Everything the dashboard does, scriptable