System Pilot — B.L.A.S.T. Protocol

Phase 1: B — Blueprint (Vision & Logic)

Discovery Questions

Ask the user all five before proceeding:

1. North Star: What is the singular desired outcome?
2. Integrations: Which external services (Slack, Shopify, Notion, etc.) are needed? Are API keys ready?
3. Source of Truth: Where does the primary data live?
4. Delivery Payload: How and where should the final result be delivered?
5. Behavioral Rules: How should the system act? (Tone, logic constraints, "Do Not" rules.)

Data-First Rule

Define the JSON Data Schema in claude.md before any code:

• What does the raw input look like?
• What does the processed output look like?

Coding only begins once the Payload shape is confirmed.

Research

Search GitHub repos and other relevant sources for libraries, patterns, or prior art that could accelerate the build. Log findings in findings.md.

Phase 2: L — Link (Connectivity)

1. Verification: Test all API connections and .env credentials first.
2. Handshake: Build minimal scripts in tools/ to verify external services are responding correctly.
3. Do not proceed to full logic if any Link is broken. Fix the connection before building on top of it.

Phase 3: A — Architect (The 3-Layer Build)

You operate within a 3-layer architecture that separates concerns to maximize reliability. LLMs are probabilistic — business logic must be deterministic.

Layer 1: Architecture (architecture/)

• Technical SOPs written in Markdown
• Define goals, inputs, tool logic, and edge cases for every component
• The Golden Rule: If logic changes, update the SOP before updating the code

Layer 2: Navigation (Decision Making)

• This is your reasoning layer
• You route data between SOPs and Tools
• You do not perform complex tasks yourself — you call execution tools in the correct order

Layer 3: Tools (tools/)

• Deterministic Python scripts — atomic and independently testable
• Environment variables and tokens stored in .env
• Use .tmp/ for all intermediate file operations
• Each script does one thing and does it reliably

Phase 4: S — Stylize (Refinement & UI)

1. Payload Refinement: Format all outputs (Slack blocks, Notion layouts, Email HTML) for professional delivery
2. UI/UX: If the project includes a dashboard or frontend, apply clean CSS/HTML and intuitive layouts
3. Feedback: Present stylized results to the user for approval before final deployment

Phase 5: T — Trigger (Deployment)

1. Cloud Transfer: Move finalized logic from local testing to the production cloud environment
2. Automation: Set up execution triggers (cron jobs, webhooks, or listeners)
3. Documentation: Finalize the Maintenance Log in claude.md for long-term stability

Operating Principles

1. Data-First Rule

Never build a Tool before the Data Schema is defined in claude.md.

After any meaningful task, update the memory files:

• progress.md — what happened, what errors occurred, what was tested
• findings.md — new discoveries, API quirks, constraints
• claude.md — update only when a schema changes, a rule is added, or architecture is modified

claude.md is law. The planning files are memory.

2. Self-Annealing (The Repair Loop)

When a tool fails or an error occurs:

1. Analyze — read the stack trace and error message. Do not guess.
2. Patch — fix the Python script in tools/
3. Test — verify the fix works
4. Update Architecture — update the corresponding .md in architecture/ with the new learning (e.g., "API requires a specific header" or "rate limit is 5 calls/sec") so the error never repeats

3. Deliverables vs. Intermediates

• Local (.tmp/) — scraped data, logs, temporary files. Ephemeral. Can be deleted.
• Global (Cloud) — the Payload. Google Sheets, databases, UI updates. A project is only "complete" when the payload is in its final cloud destination.

File Structure Reference

├── claude.md          # Project Constitution & State Tracking
├── .env               # API Keys / Secrets (verified in Link phase)
├── architecture/      # Layer 1: SOPs (the "How-To")
├── tools/             # Layer 3: Python Scripts (the "Engines")
└── .tmp/              # Temporary Workbench (Intermediates)

Session Start Checklist

Every new project session, confirm before writing any code:

• claude.md exists and has the current Data Schema
• task_plan.md has an approved Blueprint
• All API connections verified (Link phase complete)
• Discovery Questions answered by the user
• No scripts in tools/ until the above are all checked38:["$","$L40",null,{"content":"$41","frontMatter":{"name":"antigravity-blast","description":"Use this skill when building any automation, tool, script, or system in Antigravity. Trigger when the user wants to build a workflow, connect APIs, create a data pipeline, automate a process, set up a trigger or cron job, build a dashboard, or architect any multi-step system. Also trigger when the user says "build this", "automate this", "set up a system for", "create a tool that", or "I need something that does X automatically." This skill initializes the B.L.A.S.T. protocol and A.N.T. 3-layer architecture for reliable, deterministic, self-healing automation builds.\n"}}]