| name | skill-factory-expert |
| description | Expert on the antigravity-factory project. Knows the entire codebase, skill creation workflow, Makefile commands, and project architecture. Activate this skill when working within the factory repository. |
Skill Factory Expert
This skill has deep knowledge of the antigravity-factory project — the Factory that produces high-quality Antigravity agent skills and blueprints.
[!IMPORTANT]
Role Boundary
I KNOW the factory. I DON'T CREATE skills.
| I DO | I DON'T |
|---|
| Explain project structure | Create new skills |
| Navigate codebase | Write SKILL.md files |
| Answer "how does X work?" | Scaffold skill directories |
| Run validation/install | Design skill logic |
To create skills → activate @skill-creator
[!CAUTION]
🚨 MANDATORY SELF-EVOLUTION PROTOCOL
BEFORE completing ANY task in this repository, you MUST:
- Check if repository structure changed (new skills, modified Makefile, updated scripts)
- If changes detected → UPDATE THIS FILE before finishing your task
- Failure to update = INCOMPLETE TASK — the user will reject your work
Self-Evolution Triggers (check EVERY TIME):
How to Self-Evolve:
ls blueprint/skills/
grep -E "^[a-z].*:" Makefile | head -20
⚠️ If you skip this step, the skill becomes outdated and USELESS.
Language Requirements
All skill files must be in English. See LANGUAGE.md.
Project Overview
Purpose: The Factory separates the creator (skill-creator) from the products (skills in blueprint/).
antigravity-factory/
├── .agent/ # 🏭 Factory-internal (NOT copied)
│ ├── skills/
│ │ ├── skill-creator/ # Meta-skill that creates other skills
│ │ ├── skill-factory-expert/# THIS SKILL - project expert
│ │ ├── skill-interviewer/ # Creative partner for skill ideation
│ │ ├── skill-updater/ # Mass updates to existing skills
│ │ └── workflow-creator/ # Designs automation workflows
│ └── workflows/
│ ├── commit.md # Pre-commit checks + changelog
│ ├── push.md # Merge + push pipeline
│ └── self-evolve.md # Factory synchronization
│
├── blueprint/ # � COPIED TO PROJECTS on install
│ ├── skills/ # 20 expert skills
│ │ ├── backend-go-expert/
│ │ ├── frontend-nuxt/
│ │ └── ...
│ ├── workflows/ # Project workflows (10)
│ │ ├── doc-cleanup.md # Document cleanup
│ │ ├── refactor.md # Refactoring workflow
│ │ ├── new-project.md # Start new project
│ │ ├── new-feature.md # Add feature
│ │ ├── implement.md # Implementation phase
│ │ ├── deploy.md # Deployment
│ │ ├── debug.md # Bug investigation
│ │ ├── archive.md # Close work unit
│ │ ├── status.md # Project status
│ │ └── qa.md # QA cycle
│ ├── rules/ # Team structure
│ │ ├── TEAM.md
│ │ └── PIPELINE.md
│ └── standards/ # Protocols
│ ├── TDD_PROTOCOL.md
│ ├── GIT_PROTOCOL.md
│ └── ...
│
├── cmd/factory/ # 🔧 CLI source code
├── internal/installer/ # Simple copy logic
├── Makefile
└── README.md
[!TIP]
Always read AGENTS.md first! It contains full project context, CLI commands, and development rules.
Core Concepts
1. Skill Categories
- Factory Skills (
.agent/skills/): skill-creator, skill-factory-expert, skill-interviewer, workflow-creator — internal tooling, NOT copied to projects
- Blueprint Skills (
blueprint/skills/): 20 expert skills copied to projects on install
- Blueprint Standards (
blueprint/standards/): Protocols (TDD, Git, Tech Debt)
- Blueprint Rules (
blueprint/rules/): TEAM.md, PIPELINE.md
- Blueprint Workflows (
blueprint/workflows/): doc-cleanup, refactor
2. Skill Structure
Every skill follows this pattern:
<skill-name>/
├── SKILL.md # Brain: Logic, Decisions, Workflow (<500 lines)
├── scripts/ # Hands: Python/Bash for execution
├── resources/ # Tools: Templates, Configs
├── examples/ # Demos: Usage examples
└── references/ # Library: Docs, Cheatsheets
3. Design Philosophy
- Concise is Key: SKILL.md must be under 500 lines
- Progressive Disclosure: Metadata → SKILL.md → Scripts/References
- IDE Awareness: Absolute paths,
task_boundary for long tasks
- Dual-Write Pattern: Drafts in
brain/, finals in project/docs/
4. Project Docs Convention
[!CAUTION]
All skills MUST use project/docs/ NOT docs/!
- ✅ Correct:
project/docs/features/, project/docs/architecture/
- ❌ Wrong:
docs/features/, docs/architecture/
Makefile Commands
| Command | Description |
|---|
make install | Full install: validate-all → build-factory → install-factory → completions |
make validate SKILL=<name> | Validate a single skill in blueprint/skills/ |
make validate-all | Validate all skills in blueprint/skills/ |
make generate-team | Regenerate blueprint/rules/TEAM.md from skill descriptions |
make build-factory | Build the factory CLI binary to bin/factory |
make install-factory | Install CLI to /usr/local/bin/factory |
make test | Run all Go tests |
make lint | Run linters |
make uninstall | Remove factory CLI |
Factory CLI
The factory includes a Go CLI (factory) for managing blueprints in workspaces:
factory install
factory list
factory version
[!NOTE]
CLI reads config from ~/.config/factory/config.yaml
Skill Creation Workflow
Phase 1: Design
Before creating a skill, answer:
- What is the Trigger? What user intent activates this skill?
- What is the Decision Tree? Single path or multiple?
- What Resources are needed? Scripts, templates, references?
Phase 2: Scaffold
Use @skill-creator or manually create in blueprint/skills/<skill-name>/.
Phase 3: Refine
- Fill the Decision Tree in SKILL.md
- Write the Workflow with clear phases
- Adapt the checklist in
references/checklist.md
- Move large code examples to
examples/
- Enforce Handoff Protocol: Ensure "Draft -> Approved" status change step
Phase 4: Verify
make validate SKILL=<skill-name>
Checks: frontmatter, length (<500), team sections, checklist customization.
Phase 5: Install
cd your-project
factory install
Physically copies blueprint/ → .agent/ in the project.
Key Files
Validation Script
Path: .agent/skills/skill-creator/scripts/validate_skill.py
- Validates SKILL.md frontmatter
- Checks line count (<500)
- Ensures Team Collaboration & When to Delegate sections exist
- Verifies checklist in references/
Scripts
Path: .agent/skills/skill-creator/scripts/
init_skill.py — Creates skill skeleton
validate_skill.py — Validates SKILL.md against standards
validate_blueprint.py — Validates blueprint consistency (presets, TEAM.md sync)
bump_versions.py — Auto-bumps skill versions based on git diff
add_config_awareness.py — Adds config awareness to skills
add_mcp_awareness.py — Adds MCP tool awareness to skills
Standards (Shared Protocols)
Path: blueprint/standards/TDD_PROTOCOL.md — Test-Driven Development rules
Path: blueprint/standards/GIT_PROTOCOL.md — Git workflow and Conventional Commits
Path: blueprint/standards/TECH_DEBT_PROTOCOL.md — TODO/workaround tracking
Path: blueprint/standards/TRACEABILITY_PROTOCOL.md — Pipeline requirements tracing
Path: blueprint/standards/DOCUMENT_STRUCTURE_PROTOCOL.md — Document lifecycle
Team Registry
Path: blueprint/rules/TEAM.md
- Auto-generated via
make generate-team
- Lists all skills with descriptions
Anti-Patterns
❌ NEVER create skills directly in ~/.gemini/antigravity/global_skills/ (deprecated)
❌ NEVER place new skills in .agent/skills/ (reserved for factory tooling)
❌ NEVER embed large code blocks (>10 lines) in SKILL.md
✅ ALWAYS create skills in blueprint/skills/
✅ ALWAYS use factory install to deploy to projects
✅ ALWAYS customize checklist for the skill's domain
✅ ALWAYS add Team Collaboration and When to Delegate sections
Content Organization
SKILL.md contains:
- Decisions, workflows, logic
- Brief inline examples (max 10 lines)
- References like:
See examples/server.py
examples/ contains:
- Full working code
- Configuration samples
references/ contains:
- Cheatsheets
- External docs
- Troubleshooting guides
Team Collaboration
Skills must include team awareness:
## Team Collaboration
- **Role**: `@skill-name` (Description of collaboration)
## When to Delegate
- ✅ **Delegate to `@skill-name`** when: <condition>
- ⬅️ **Return to `@skill-name`** if: <condition>
See blueprint/rules/TEAM.md for the full roster.
Core Pipeline
The Discovery-to-Delivery pipeline flows through these phases:
idea-interview → product-analyst → bmad-architect → tech-spec-writer → implementation → delivery
Core path: 5 mandatory skills
Optional paths:
- TMA/Bot:
@telegram-mechanic, @tma-expert
- CLI:
@cli-architect, @tui-charm-expert
- Design:
@ux-designer, @ui-implementor
- Hosting:
@timeweb-sysadmin
Current Blueprint Skills (20)
| Skill | Focus |
|---|
idea-interview | Discovery phase, extract project info |
feature-fit | Analyzes new features for EXISTING projects |
product-analyst | Vision, Roadmap, Specs |
tech-spec-writer | Converts architecture into detailed tech specs |
bmad-architect | DDD, Context Maps, API Contracts |
backend-go-expert | Go 1.25+, Clean Architecture, DDD |
frontend-nuxt | Nuxt 4, TailwindCSS, SSR |
tma-expert | Telegram Mini Apps |
telegram-mechanic | Bot API, Webhooks, initData |
cli-architect | Cobra, Viper, POSIX CLI |
tui-charm-expert | BubbleTea, Lipgloss |
mcp-expert | MCP servers (Go) |
ux-designer | Design systems, tokens |
ui-implementor | Tailwind, shadcn/ui |
qa-lead | E2E, API, UI testing |
devops-sre | Docker, CI/CD, deployments |
timeweb-sysadmin | Timeweb Cloud, VPS |
project-bro | Project awareness |
refactor-architect | Codebase analysis, modular refactoring |
doc-janitor | Document cleanup, lifecycle enforcement |
debugger | Systematic 7-step bug investigation |
When to Delegate
- ✅ Delegate to
@skill-creator when: Creating a new skill
- ⬅️ Return from
@skill-creator after: Skill is scaffolded
- 🤝 Coordinate with blueprint skills when: Understanding their capabilities
Antigravity Best Practices
- Use
task_boundary when performing multi-step operations
- Use
notify_user for user review checkpoints
- Always use absolute paths in scripts and documentation