Welcome to Antigravity Prism. This repository implements a Google L7-Quality Autonomous Software Organization. It is not just code; it is a system of specialized agents, contracts, and workflows that function as a mature engineering team.
The Prism Protocol operates in hybrid mode:
- Antigravity (Claude.ai): Planning, design, spec generation, and auditing
- Claude Code (Terminal): Code execution, testing, and implementation
/product → [TASKNAME]_PRD.md → /devteam → [TASKNAME]_specs.md → /software_engineer → TERMINAL PROMPT
Every task begins by establishing a Task Name in the Product phase:
/product Add user authentication for the API
The agent will:
- Establish
[TASKNAME](e.g.,USER_AUTH) - Create an Implementation Plan
- Generate
docs/USER_AUTH_PRD.md - Provide handoff: "Invoke
/devteam USER_AUTHto proceed."
/devteam USER_AUTH
The agent will:
- Validate
docs/USER_AUTH_PRD.mdexists - Create
specs/USER_AUTH_specs.md - Provide handoff: "Invoke
/software_engineer USER_AUTHto proceed."
/software_engineer USER_AUTH
The agent will:
- Run L7 Spec Review Gate (Pre-Mortem, Antagonist, Complexity)
- Generate a TERMINAL PROMPT with L7 recursive audits
- Include Zero-Trust Review before commit instructions
Copy the TERMINAL PROMPT and run in Claude Code:
claude -p "[Paste TERMINAL PROMPT here]"Paste the terminal output back into Antigravity for the Audit phase. On pass, invoke /NEXT to finalize.
- Mission: Discovery, Strategy, Definition
- Use when: You have an idea, problem, or business goal
- Output:
docs/[TASKNAME]_PRD.md
- Mission: Architecture, Engineering, Delivery
- Use when: You have a PRD and need technical specs
- Output:
specs/[TASKNAME]_specs.md
- Mission: Context preparation, L7 review, prompt generation
- Use when: You have specs and need executable instructions
- Output: TERMINAL PROMPT for Claude Code
- Mission: Reliability, Security, Support
- Use when: Something is broken, slow, or needs maintenance
- Output: Stability, Incident Resolution
All tasks use SCREAMING_SNAKE_CASE:
| User Request | Task Name |
|---|---|
| "Add user login" | USER_LOGIN |
| "Implement JWT authentication" | JWT_AUTH |
| "Dashboard analytics feature" | DASHBOARD_ANALYTICS |
Before generating any terminal prompt:
- Pre-Mortem: "If this fails in production, what was the cause?"
- Antagonist: "How would a malicious actor exploit this?"
- Complexity: "Is this over-engineered?"
Before any commit instruction:
- Security Checklist (no secrets, input validation, auth checks)
- Quality Checklist (tests pass, no linter errors)
- Architecture Checklist (follows patterns, no breaking changes)
Contracts are located in specs/templates/:
- PRODUCT_SPEC.md: Requirements contract
- TECH_DESIGN_SPEC.md: Architecture contract
- OPS_READINESS_SPEC.md: Deployment contract
Project management via planning/:
- ROADMAP.md: Quarterly Goals
- BACKLOG.md: Prioritized work
- SPRINT_BOARD.md: Active execution
- User Guide: Detailed playbooks for common scenarios