antigravity-plan
/antigravity-plan
Pisze plany wdrożeń w formacie implementation_plan.md z celem, otwartymi pytaniami, zmianami według plików i planem weryfikacji.
- Autor
- jshsinaga ↗
- Repozytorium
- antigravity-plan-skill ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 31.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 10
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 31.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (10). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Pisze plany wdrożeń w formacie implementation_plan.md z celem, otwartymi pytaniami, zmianami według plików i planem weryfikacji.
Wyzwalacze z opisu
Oryginalny opis (EN)
Write implementation plans in the Antigravity `implementation_plan.md` format (goal description, User Review Required, Open Questions, Proposed Changes with [NEW]/[MODIFY]/[DELETE] file entries, Verification Plan). Use this skill whenever the user asks for a plan, implementation plan, design doc, or "plan this before coding" for a feature, refactor, bugfix, or migration, in any language, or mentions Antigravity-style plans, even if they don't name the format. Respond with the full Markdown plan in chat and wait for approval before coding. Save a plan file only when the user explicitly requests
Jak działa
- Workflow
- Language
- Template
- Rules for each part
- Quality checklist
- Porting to other tools
Podgląd SKILL.md
---
name: antigravity-plan
license: MIT
description: Write implementation plans in the Antigravity `implementation_plan.md` format (goal description, User Review Required, Open Questions, Proposed Changes with [NEW]/[MODIFY]/[DELETE] file entries, Verification Plan). Use this skill whenever the user asks for a plan, implementation plan, design doc, or "plan this before coding" for a feature, refactor, bugfix, or migration, in any language, or mentions Antigravity-style plans, even if they don't name the format. Respond with the full Markdown plan in chat and wait for approval before coding. Save a plan file only when the user explicitly requests it.
---
# Antigravity Plan
Produce an implementation plan in the structure used by Google Antigravity's planning mode. The format works because it separates *what needs a human decision* (review and open questions) from *what will change* (per-file entries) from *how to prove it works* (verification). A reader can approve or reject the plan in two minutes without reading code.
## Workflow
1. **Explore first.** Read the relevant code, schema, and existing tests before writing. A plan that names real files, functions, and constraints is useful; a generic one is not.
2. **Respond with the plan in chat.** Use the template below for the next user-facing planning response. Render the full plan as Markdown, including its headings, alerts, and file entries; do not wrap the whole plan in a code fence or substitute a file path or summary for it.
3. **Save only on explicit request.** Planning requests authorize a chat response, not file creation or updates. Create or update a plan file only if the user explicitly asks to save, export, or edit one. Honor their requested path; if they request a file without specifying a path, use `implementation_plan.md` in the project root, or `docs/plans/<slug>.md` if the repo already uses that plans directory. A writable workspace, an existing plan file, or a tool's artifact convention is not a request to write a file. Keep the full plan in chat even when saving is requested.
4. **Stop and ask for review.** Do not edit source files until the user approves. If the user answers Open Questions or requests changes, present the revised plan in chat. If they requested a saved plan, update that same file too.
5. After approval, implement section by section following `Proposed Changes`, then run the `Verification Plan`.
## Language
This skill file is written in English, but the plan it produces must be written in the user's language (the language of their request, unless they ask for another).
- Write all prose, alert text, bullets, and manual verification steps in the user's language.
- Keep these four anchor headings in English in every language, because tools and readers rely on them: `User Review Required`, `Open Questions`, `Proposed Changes`, `Verification Plan`.
- Translate every other heading (the title, optional sections, component names, and the Automated/Manual sub-headings) into the user's language.
- Keep code, file paths, commands, and identifiers exactly as they are in the project.
## Template
Use this structure, in this order. Separate major sections with `---`. Bracketed text is a placeholder to replace, and headings marked "translate" follow the Language rules above.
````markdown
# [Plan Title] <!-- translate -->
[1-3 sentences: what is being built or fixed, and why. State the outcome, not the process.]
---
## Problem Description & Goals <!-- optional, translate -->
<!-- Include for non-trivial or bug-driven changes; omit for small ones. -->
### Background <!-- translate -->
[Numbered list of the concrete causes or gaps in the current system.]
### Goals <!-- translate -->
[Numbered list of what will be true after the change.]
---
## User Review Required
> [!IMPORTANT]
> **[Decision or definition needing approval]**
> [Rules, formulas, or behavior the user must confirm before work starts.]
> [!WARNING]
> **[Destructive or risky action]**
> [What is lost or hard to undo, and the safeguard.]
> [!NOTE]
> **[Compatibility or migration info]**
> [Why existing data or behavior stays safe, e.g. no DB migration needed.]
## Open Questions
[Blocking questions for the user, one per bullet. If none, write one line saying there are no blocking questions and why.]
---
## [Flow / Architecture] <!-- optional, translate -->
```mermaid
flowchart TD
A[Step] --> B[Step]
```
---
## Proposed Changes
### 1. [Component or Layer Name] <!-- translate -->
#### [MODIFY] [`path/to/file.ext`](file:///absolute/path/to/file.ext)
- [What changes, at behavior level. Name the functions, types, or queries involved.]
- [Next change in this file.]
```lang
// Only for contracts that are easy to get wrong: signatures, SQL, route registration, types.
```
#### [NEW] [`path/to/new-file.ext`](file:///absolute/path/to/new-file.ext)
- [Purpose of the file and its public surface.]
#### [DELETE] [`path/to/old-file.ext`](file:///absolute/path/to/old-file.ext)
- [Why it is safe to remove.]
---
### 2. [Next Component or Layer Name]
...
---
## Verification Plan
### Automated Tests <!-- translate -->
[Exact commands in fenced code blocks, then the test files to add or update, one line each saying what each covers.]
### Manual Verification <!-- translate -->
1. [Concrete step the user can perform and the exact result to expect.]
2. ...
````
The `<!-- ... -->` comments are guidance for you; do not copy them into the plan.
## Rules for each part
**Title and summary.** Title follows the pattern "Implementation Plan: <Feature>" (translated). The summary says what and why in plain words.
**User Review Required.** Use GitHub alert blockquotes. Each alert has a bold title line, then the body. Pick the type by meaning:
- `[!IMPORTANT]`: a definition, formula, or design choice that changes the result if the user disagrees.
- `[!WARNING]`: data loss, irreversible operations, security or integrity risk.
- `[!NOTE]`: reassurance or context (backward compatibility, no migration).
Include only alerts that carry a real decision or risk. An empty or filler alert trains reviewers to skip the section. The alert markers themselves (`[!IMPORTANT]` etc.) stay in English.
**Open Questions.** Only questions that block the plan. If the user already chose an approach, say so instead of reopening it.
**Diagram.** Prefer `sequenceDiagram` for request and transaction flows and `flowchart TD` for data or component flow. Keep node labels short. If the target tool cannot render mermaid, keep the block anyway; it still reads as structured text.
**Proposed Changes.**
- Group by layer or component (data model, store, service, handler/router, API contract, UI, tests), numbered `### 1.`, `### 2.`, in dependency order: lower layers first, so each section can build on the previous one.
- Each file gets a `####` header: `[MODIFY]`, `[NEW]`, or `[DELETE]` (always in English), then a markdown link whose label is the path in backticks.
- Link target is `file:///` plus the absolute path with forward slashes (for example `file:///d:/PROJECTS/app/src/lib/song.ts`). If the absolute path is unknown, or the tool is not Antigravity, use the repo-relative path in backticks without a link; do not invent absolute paths.
- Under each file, describe the change as short bullets. Reference real symbol names. Add a code block only when the exact signature, SQL, route, or type matters; never paste full implementations.
- Put test files in `Proposed Changes` only when the plan needs to explain them in detail; otherwise list them under `Verification Plan`.
**Verification Plan.**
- Automated tests list runnable commands exactly as they would be typed (`npm test`, `go test ./...`, lint, typecheck, build), plus new or updated test files with one line on what each checks.
- Manual verification is a numbered, user-executable script: where to click or call, and what must be observed. Cover the happy path, the cancel or negative path, and any destructive step.
## Quality checklist
Before presenting the plan, check:
- The full plan is the chat response, and any plan file creation or update was explicitly requested by the user.
- Every file path comes from code you actually inspected.
- Every risky or irreversible step has a `[!WARNING]` and a confirmation or rollback story.
- Each Proposed Changes section can be implemented without reading the others out of order.
- Verification commands exist in the project (check `package.json`, `Makefile`, CI config).
- The plan's prose and translatable headings are in the user's language, and the four anchor headings are in English.
- The plan is as long as the change requires and no longer: a one-file bugfix needs a short plan with only the sections that apply.
## Porting to other tools
The format is plain markdown, so it transfers as-is. Adjust only these when the host tool differs:
- Apply the chat response and explicit-save rules from Workflow in every host tool.
- Replace `file:///` links with repo-relative backticked paths if links do not resolve.
- If the tool renders GitHub alerts poorly, the `> [!TYPE]` lines still read as labeled blockquotes; leave them in place.
Wersje
W bazie nie ma innych skilli o nazwie antigravity-plan.
Migawka bazy z 13.11.2026 · wersja paczki 407d81860028 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 95write-implementation-plan
Przygotowuje plan wdrożenia po zbadaniu kodu: listę plików do zmiany, otwarte pytania i plan weryfikacji. - 100writing-plans
Pisze uporządkowane plany wdrożenia na podstawie specyfikacji lub wymagań, zanim ktokolwiek zmieni kod. - 100how-to-plan
Pisze plany gotowe do przekazania: cel do 150 słów, atomowe kroki commitów z bramkami weryfikacji, drzewa plików przed i po oraz dwukluczowe potwierdzenie przed usunięciem planu. - 100dev-plan
Tworzy konkretne plany implementacji po poznaniu wymagań: kroki na poziomie plików, kolejność, strategię testów, ryzyka i weryfikację przed kodowaniem funkcji, poprawki, refaktoryzacji lub migracji. - 100execute-plan
Wykonuje ustrukturyzowane plany zadań z plików w katalogu plans i śledzi status realizacji poszczególnych kroków. - 100writing-a-design-doc
Tworzy i edytuje dokumenty projektowe Archdocs z diagramami, specyfikacją, planem wdrożenia i otwartymi pytaniami.