/implement orchestrates feature implementation task by task, following the plan produced by /plan. Each task goes through a confidence check, a full TDD cycle, a commit, and a code review before moving to the next task. The loop continues until all plan tasks are complete.
Prerequisites
Before starting/implement, all four conditions must be met:
- Approved design doc exists at
docs/plans/*-design.md— if not, run/brainstormfirst - Implementation plan exists at
docs/plans/*-plan.md— if not, run/planfirst - You are on a feature branch (not
mainormaster) - Fast Mode is active in your Antigravity settings
Session state
At the start of each task, SuperAntigravity outputs its current state:Orchestration flow
1
Load the plan
SuperAntigravity reads the plan file at
docs/plans/*-plan.md before running any confidence check. A confidence check that doesn’t reference the plan’s specific tasks and file paths is invalid.2
Confidence check (score ≥ 27 required)
SuperAntigravity runs the confidence-check skill against the current task. A score of 27 or higher out of 30 is required to proceed. If confidence is below 27, see the section below on what happens.
3
RED phase — write the failing test
The test-driven-development skill writes one minimal test for the behavior being implemented. The test must be run and confirmed to fail before any implementation code is written.
4
Minimal implementation
Writes the simplest possible code to make the failing test pass. Nothing beyond what the test requires.
5
GREEN phase — verify the test passes
Runs the test and confirms it passes. Also confirms no other tests have broken. Output must be pristine — no errors or warnings.
6
Commit
Commits immediately after the GREEN phase. Message format:
test: [what behavior is now tested]. Never accumulates more than one GREEN test before committing.7
Code review
Loads the requesting-code-review skill to review the task’s implementation against the plan. Identified issues are addressed before moving to the next task.
8
Next task (repeat)
Returns to the confidence check for the next task. The loop continues until all plan tasks are complete.
9
Verification before completion
After all tasks are done, loads the verification-before-completion skill to confirm the full implementation matches the plan’s requirements.
10
Finishing the branch
Loads the finishing-a-development-branch skill to present the merge or PR options.
Commit cadence
Super Antigravity commits after every GREEN test — no exceptions.When confidence is below 27
If the confidence check returns a score below 27, SuperAntigravity announces:When a test fails after implementation
If a test fails after implementation code has been written, SuperAntigravity loads the systematic-debugging skill. It does not retry without diagnosis. The debugging skill identifies the root cause before any fix is attempted.Skills used
Interrupt protocols
Bug discovered mid-task. Systematic-debugging is loaded. The bug is fixed and a regression test is added. Then the plan resumes from where it left off — the current task is not abandoned. Scope change requested. Implementation stops. You are directed to/brainstorm for the new scope. The existing plan is not modified without a new approved design. Scope changes are never silently absorbed.