When this skill fires
The skill description reads: “Use when you have a spec or requirements for a multi-step task, before touching code.” It fires automatically at the end of the brainstorming skill, and also when you start a session with an existing spec or design document ready to implement.What it does
Writing-plans produces a structured markdown document saved todocs/plans/YYYY-MM-DD-<feature-name>.md. The document is a complete, self-contained guide for implementing the feature using TDD. Someone who has never seen the codebase should be able to execute it without asking a single question.
The skill announces its activation:
“I’m using the writing-plans skill to create the implementation plan.”
How it works
Plan structure
Every plan starts with a mandatory header:Bite-sized task granularity
The core rule: each step produces exactly one artifact. If a step produces two artifacts, split it.| Artifact | Example step |
|---|---|
| Test file | ”Write the failing test” |
| Test run output | ”Run it to make sure it fails” |
| Code file | ”Implement the minimal code to make the test pass” |
| Passing test output | ”Run the tests and make sure they pass” |
| Commit | ”Commit” |
Task format
Each task in the plan follows this structure:Write the failing test
Include the complete test code — not a description of what to write, but the actual test.
Run test to verify it fails
Include the exact command and the expected failure output. For example:
pytest tests/path/test.py::test_name -v → Expected: FAIL with “function not defined”.Write minimal implementation
Include the complete implementation code. No placeholders, no “add validation here”.
Plan validation
Before handing the plan to execution, the skill reads it from the perspective of an engineer who has never seen the codebase. The validation checklist:- Every file path is exact and follows the documented convention
- Every command includes expected output
- Every task’s purpose is clear from its name and description alone
- No task assumes knowledge of a prior conversation or undocumented context
Execution handoff
After saving the plan, the skill offers two execution options: Option 1 — Subagent-driven (this session): Fresh subagent per task, code review between tasks, fast iteration. Requires switching Antigravity to Fast Mode. Option 2 — Parallel session (separate): Open a new session with the executing-plans skill, batch execution with checkpoints. Both options load theexecuting-plans skill.
Example scenario
You have an approved design for a notifications feature. The writing-plans skill:- Creates
docs/plans/2026-03-17-notifications.md - Writes Task 1: create the
notificationstable migration with failing test, run command, implementation, passing run, commit - Writes Task 2: create the
NotificationWorkerclass with failing test, implementation, passing run, commit - Validates each task has enough context to execute without questions
- Presents two execution options and waits for your choice
Related skills
Brainstorming
Produces the approved design that writing-plans converts into tasks.
Executing plans
Consumes the plan document and implements it in batches.
Test-driven development
The TDD cycle that every task in the plan follows.