| name | sprint-loops |
| description | Structured five-phase workflow (Research → Plan → Build → Test → Loop) that translates Antigravity's native Planning Mode into the Sprint Loops cross-harness schema. |
Sprint Loops — Antigravity Translation Layer
You are working in a Sprint Loop. Each sprint is a five-phase sequence: Research → Plan → Build → Test → Loop.
In Antigravity IDE, we execute this by using Antigravity's native Planning Mode (implementation_plan.md, task.md, walkthrough.md) as the engine, and syncing that state to the Sprint Loops filesystem schema (sprints/sN/, agent-tasks/). This filesystem schema acts as a Rosetta Stone for cross-harness compatibility, ensuring other agents (like Claude Code or Codex) can seamlessly pick up where you left off.
Phase Playbook
Determine the current phase by inspecting the filesystem (or running sh antigravity-ide/skills/sprint-loop/scripts/current-phase.sh). Follow the instructions for your current phase.
Phase 1: Initialize
- If
sprints/ does not exist, initialize sprints/s0/. You can use sh antigravity-ide/skills/sprint-loop/scripts/init-sprint.sh or manually create the directory structure defined in the Sprint Loops schema.
- Ensure
agent-tasks/agent-tasks.md and agent-tasks/completed-tasks.md exist.
- Write
sprints/s0/sprint-meta.md.
Phase 2: Research
- Use Antigravity's native Research capabilities to explore the codebase.
- Record findings, open questions, and external research in
sprints/sN/sprint-research/research-report.md (use schemas/research-report.md as a guide).
- Conclude by entering the Plan Phase.
Phase 3: Plan
- Generate Antigravity's native
implementation_plan.md artifact to present to the user.
- The plan should lay out: "For this phase we're focusing on X, so I'm going to begin by researching the internal codebase and test results, then I am going to record any open questions and do external research to confirm what I think to be true and answer open questions. Then I will write the build plan. Then I will execute the build plan, then I will design the tests based on what was built, emphasizing the sprint-loops testing schema of expanding from localized mocks and stubs gradually to e2e testing as much as I can, documenting everything in the test results."
- SYNC STEP: Once the user approves
implementation_plan.md, you MUST copy/sync the relevant build steps into sprints/sN/sprint-plans/build-plan.md and the test plan into sprints/sN/sprint-plans/test-plan.md.
- Add
Finalized - DO NOT EDIT at the top of build-plan.md and test-plan.md.
- Populate
agent-tasks/agent-tasks.md with the items from the build plan.
Phase 4: Build
- Use Antigravity's native
task.md artifact as your active checklist to execute the build-plan.md.
- SYNC STEP: As you complete tasks in
task.md, you MUST sync them by removing them from agent-tasks/agent-tasks.md and appending them to agent-tasks/completed-tasks.md.
- Make an atomic git commit for each completed task using
sh antigravity-ide/skills/sprint-loop/scripts/commit-task.sh (or standard git commit following the same format).
Phase 5: Test
- Design and execute tests based on what was built.
- Emphasize the Sprint Loops testing schema: expanding from localized mocks and stubs gradually to E2E testing as much as possible.
- If the project defines a canonical suite runner (e.g. a
tools/run-guards.sh), invoke that runner rather than ad-hoc commands and record its confirmations plus the CI conclusion (head SHA, run URL) in the test-report's CI Confirmation block — or "CI not configured — local confirmations only".
- SYNC STEP: Document everything in
sprints/sN/sprint-tests/test-report.md (or failure-report.md if blocked).
Phase 6: Loop
- Create Antigravity's native
walkthrough.md artifact to present your final work to the user.
- Update
sprints/sN/sprint-meta.md to set status to completed (or failed if irrecoverable).
- Verify a clean working tree.
- Inform the user that the sprint is closed and they can start the next one.
Core Rule: Never skip the SYNC STEPS. Antigravity's internal artifacts are your tools, but the Sprint Loops sprints/ and agent-tasks/ directories are the authoritative project state.