๐ŸŽ–๏ธ
DoW SkillBridge & Veterans

Free Tech Training for Military Veterans

Microsoft MSSA, Google Cloud Launchpad, AWS Educate, Salesforce VetForce, Onward to Opportunity, CISA Learning, ChatGPT Plus free, SANS VetSuccess, and more. Over $50,000+ in combined free training for veterans, service members, and spouses. LinkedIn Premium = 1 free year for veterans.

All programs FREE | 3,000+ SkillBridge partners | 2026: AF/SF/Army added rank-based duration caps | Updated May 12, 2026

๐Ÿš€
Updated May 12, 2026

What's New in AI & Cybersecurity

Latest model releases, free offers, and Sports Media updates - all live right now.

Sports Media SOPMay 7, 2026

AI-Driven Research SOP (5 stages)

Perplexity, Gemini, DeepSeek, Claude, then human audit. DeepSeek approved for non-sensitive research via Western-hosted endpoints only.

Read the SOP โ†’
AnthropicMay 10, 2026

Claude Opus 4.7 GA + doubled rate limits

Opus 4.7 reaches GA with improved coding and higher-resolution vision. Claude Code 5-hour rate limits doubled for Pro / Max / Team thanks to the SpaceX Colossus 1 compute deal.

Read announcement โ†’
AnthropicMay 11, 2026

Agent View (Claude Code v2.1.139+)

New unified dashboard for parallel sessions. Spawn sessions, see state (working / waiting / done), and attach with Alt+1..9. Open via `claude agents` or โ† from any session.

Read announcement โ†’
CybersecurityMay 8, 2026

Canvas LMS extortion + Apache HTTP/2 + Linux LPE

Canvas LMS extortion threat hits ~9,000 institutions / 275M records. Apache HTTP/2 CVE-2026-23918 + a Linux LPE (CVE-2026-31431) added to CISA KEV. Patch and review your SIEM rules.

Read the brief โ†’
DeepSeekApr 24, 2026

DeepSeek-V4 preview (open-weights)

V4 Flash + V4 Pro MoE, 1M-token context, MIT license. Strong on coding (SWE-bench 80.6) and reasoning. Use Together.ai / Fireworks / OpenRouter for company work - never .cn endpoints.

Read announcement โ†’
OpenAIMay 5, 2026

GPT-5.5 Instant is the new default

52.5% fewer hallucinations on high-stakes prompts; new Memory Sources (search past chats, files, Gmail). GPT-5.3 Instant retained 3 months.

Read announcement โ†’
xAIApr 30, 2026

Grok 4.3 (Grok 5 still pending)

Grok 4.3 brings 1M context, native video input, and file generation. Grok 5 public beta now expected Mayโ€“June 2026.

View news โ†’
GoogleFeb 19, 2026

Gemini 3.1 Pro is current

Top-of-the-line reasoning (ARC-AGI-2 77.1%). Free 12-month student offer CLOSED Mar 11, 2026 - free trial + AI Studio still open.

Read announcement โ†’
Sports MediaApr 2026

New: Cybersecurity Hub + Passbolt

Our team password manager, plus 30+ free cybersecurity tools, training (TryHackMe, HackTheBox), and certs (ISC2 CC, Cisco Veterans).

Open Cybersecurity Hub โ†’
PerplexityApr 2026

Comet Browser is now FREE

Comet (iOS, Android, Windows, Mac) is free for everyone. Bundles GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro into one assistant.

Get Comet โ†’
GitHubMar 12, 2026

Copilot Student plan launched

New Copilot Student plan with unlimited completions + premium models + premium-request allowance. NEW SIGN-UPS PAUSED Apr 20, 2026 - check back soon.

GitHub Student Pack โ†’
Sports MediaApr 29, 2026

DoW SkillBridge + AI for Vets refresh

New AWS Educate + Hiring Our Heroes + ChatGPT Plus Free for Vets sections. AF/SF/Army added rank-based duration caps in 2026.

Open Veterans Hub โ†’
Linux FoundationNov 25, 2025

MCP spec 2025-11-25 + AAIF

Latest MCP spec adds OAuth 2.1 + PKCE + Resource Indicators. Anthropic donated MCP to the LF Agentic AI Foundation in Dec 2025.

Read spec โ†’
๐ŸŽ–๏ธ
Veterans - Free 1 Year

OpenAI ChatGPT Plus - FREE for Veterans ($240 Value)

1 full year of ChatGPT Plus free for transitioning servicemembers (within 12 months of separation) and recently separated veterans. Includes GPT-5.5 (released April 23, 2026), advanced voice, image generation, Sora 2 video, Agent Mode, and 100 military transition prompts.

Active duty within 12 months of separation + recently separated veterans | Verify at chatgpt.com

๐Ÿ›ก๏ธ
Military & Gov - Free 1 Year

Perplexity Pro - FREE for .mil & .gov Users

1 full year of Perplexity Pro free for all .mil and .gov email holders. Automatic zero-data-retention security. No code needed - banner appears automatically on homepage. AI-powered search with citations.

.mil and .gov email holders | Automatic activation | Zero-data-retention security

๐Ÿง 
Now Available

Claude AI by Anthropic - Fully Approved

Claude is now fully approved and available for all Sports Media projects. Access Claude Opus 4.7 (launched April 16, 2026), Sonnet 4.6, Haiku 4.5, Claude Code CLI v2.1+, Agent Teams, Extended Thinking, and Deep Research. Free tier + 30-day Pro trial available.

Opus 4.7 (NEW) | Free tier | 30-day Pro trial | Student Builders: $50 free API credits | Campus Spring 2026 closed (Fall cohort coming)

๐ŸŽ“
Free for Students

Google Gemini 3.1 Pro - FREE for 1 Year

The 12-month Google AI Pro (Gemini 3) free student offer closed March 11, 2026. A free 1-month Google AI Pro trial is still available, plus existing enrollees may renew. Continuing students should also check Google Workspace for Education access via their .edu domain.

Free 1-month trial still open | Education Pro discounts via .edu domain | Renewals available for prior enrollees

โšก
Free for Students

Cursor AI Pro - FREE for 1 Year ($240 Value)

Students get Cursor Pro free for 12 months. AI-powered code editor with GPT-4o, Gemini, Claude, 500 fast requests/month, Tab Autocomplete, and codebase chat.

Verify with .edu email or GitHub Student Developer Pack | Works worldwide

๐Ÿ”ฎ
Students - 12 Months Free

Perplexity Pro - FREE 12 Months for Students

Students get 12 months of Perplexity Pro free with .edu email verification. No credit card required. Plus earn up to 24 additional months free through referrals. After free year: 50% discount at $10/month.

Referral program ends May 31, 2026 | Verify with .edu email | No credit card needed

๐Ÿ’ป
New - Jan 2026

Microsoft 365 Premium + LinkedIn Premium - FREE 12 Months

U.S. college students get Microsoft 365 Premium (Word, Excel, PowerPoint, Outlook with Copilot AI) + LinkedIn Premium Career (24K+ courses, InMail, salary insights) free for 12 months.

US college students only | Verify with .edu email | Announced January 2026

๐ŸŽ–๏ธ
Veterans - Free AI Training

Google Launchpad - Free Gen AI Training & Certification

3-week free virtual gen AI training for US/Canada veterans. Graduates receive Google's Gen AI Leader certification. 4,000+ veterans already trained.

US & Canada veterans | No cost | Google Gen AI Leader certification included

๐Ÿ™
New Sign-ups Paused (Apr 20, 2026)

GitHub Copilot Student Plan - Launched March 12, 2026

Copilot Student launched March 12, 2026 with unlimited completions, premium models, and a monthly premium-request allowance for verified students. New sign-ups were temporarily paused on April 20, 2026 - existing users keep their access; check back for re-open.

Existing users keep access | New sign-ups paused | Pair with rest of GitHub Student Pack ($500+ value)

๐Ÿ›ก๏ธ
Last Call - Closes May 20, 2026

ISC2 Certified in Cybersecurity (CC) - FREE Course + Exam

Free online course AND free exam through ISC2โ€™s One Million Certified in Cybersecurity program. Industry-recognized entry-level cyber cert. New enrollments close May 20, 2026; exam codes must be used by Dec 31, 2026.

Anyone eligible | $0 course + $0 exam | Sign up immediately to lock in your code

๐ŸŽ–๏ธ
Veterans & Military - Free

Cisco Veteran Cyber Scholarship - Free CCNA / CyberOps

~255 hours of free Cisco cybersecurity training + free certification exam vouchers for veterans, transitioning servicemembers, Guard, Reserve, and military spouses. Choose CCNA, CyberOps Associate, or CCNA Security tracks.

US veterans & military | Free training + free exam | Pair with SANS VetSuccess for GIAC certs

๐Ÿ”
Sports Media Standard - Required

Passbolt - Our Team Password Manager

Open-source, end-to-end encrypted password manager used by Sports Media Inc. for all team credentials. Available on Web, Chrome, Firefox, Windows, macOS, Linux, iOS, and Android. Required for every intern before requesting any company credential.

Free Community Edition | OpenPGP E2E encryption | All platforms supported

๐Ÿ“š

Over $3,500 in Free Student Programs Available

M365 Premium, Azure, JetBrains, Figma, Cursor AI, Claude, Notion, Copilot Pro, GitHub Pack, and 35+ more verified programs

View All Programs
๐Ÿ›ก๏ธ
Sports Media Standard

Passbolt - Our Team Password Manager

Open-source, end-to-end encrypted password manager. Required for every intern and engineer before requesting any company credential.

Sports Media Intern Tooling Disclaimer

Sports Media interns are not required to buy anything (no paid accounts, no subscriptions) to participate in the internship or complete assigned work.

Sports Media will provide all necessary tools to get projects completed, including access to AI tooling as needed.

Standard Operating Procedure (SOP) for Tooling:

  • Default for interns: Use ChatGPT on our corporate Sports Media Inc. OpenAI account for coding and project work. Perplexity Pro (company-provided) is also available.
  • Interns should use the Perplexity account provided by leadership and must not share the credentials.
  • Interns are allowed but not required to use their own tools:
    • Their own ChatGPT / OpenAI account (Preferred)
    • Their own Gemini 3 account or GitHub Copilot
    • Their own Claude account (Anthropic)
    • Any other AI assistant they personally have

Interns are NOT required to purchase ChatGPT, Claude, Gemini, or any other paid AI tool for this internship. All major AI platforms are approved for use.

Access Instructions:

If you do not yet have access to a company-provided Perplexity Pro account, contact your Team Lead or appropriate leadership via Slack channels like #5-training-new-interns or #computer-science to get account details and login instructions.

Student Programs Available:

Students can also get their own free Perplexity Pro account through the Perplexity Student Program. Check out all 30+ free student programs including Azure ($100 credit), JetBrains IDEs, Figma Pro, and more!

Security Note:

  • Do NOT paste passwords, API keys, private keys, or other secrets into AI prompts.
  • Always use environment variables, secrets managers, or encrypted/permissioned documents for sensitive data.

Tooling SOP for Interns

Preferred Workflow

  • Using company-provided tools: Use ChatGPT on our corporate Sports Media Inc. OpenAI account for coding, documentation, and project work.
  • Using personal accounts: If you already have your own accounts, you may use ChatGPT, Claude, Gemini 3, or other approved AI accounts.

Training Focus

This site's training content focuses on:

  • ChatGPT and AI tool usage patterns
  • Proper use of PRDs (Product Requirements Documents)
  • Creating and maintaining CLAUDE.md project context files
  • Following company SOPs for development workflows

Deep Training Available: Sports Media provides comprehensive AI training for all interns and staff. Access our full training curriculum - your email is already attached from joining our intern program.

Handoff Expectations

Regardless of which AI tool you use, your outputs must be clear enough that any Sports Media engineer can pick up the project and continue. This means:

  • Clean, well-documented code
  • Up-to-date CLAUDE.md file with project context
  • PRDs that define scope, user stories, and acceptance criteria
  • Clear commit messages and PR descriptions

AI Tools Overview

๐Ÿง 

Claude

Anthropic's AI Assistant

  • โ€ข Opus 4.7 (NEW Apr 2026) - Best for coding
  • โ€ข Sonnet 4.6 - Best for research & everyday use
  • โ€ข Haiku 4.5 - Fastest & cheapest
  • โ€ข Extended thinking & deep research
  • โ€ข Claude Code CLI v2.1+ for terminal
๐Ÿ”ต

Gemini 3.1 Pro

Google's AI Platform

  • โ€ข Gemini 3.1 Pro (top reasoning)
  • โ€ข Gemini 3 Flash (fast, default)
  • โ€ข Gemini CLI + Code Assist in VS Code
  • โ€ข AI Studio free tier (no card required)
๐ŸŸข

ChatGPT (GPT-5.5)

OpenAI's Assistant

  • โ€ข GPT-5.5 Instant (May 5, 2026) default
  • โ€ข GPT-5.5 Pro for hardest tasks
  • โ€ข OpenAI CLI + GitHub Copilot in VS Code
  • โ€ข Sora 2 video + Agent Mode + Memory Sources
๐Ÿ‹

DeepSeek V4

Open-Weight Reasoning

  • โ€ข V4 Flash + V4 Pro (Apr 24, 2026) MoE, 1M context
  • โ€ข R1 reasoning model still available
  • โ€ข MIT-licensed weights - self-hostable
  • โ€ข Free chat at chat.deepseek.com (no Pro tier)
  • โš  Western-host only for company data
๐Ÿ›ก๏ธ

Approved AI Platforms

Sports Media Inc. approves Claude (Anthropic), ChatGPT (OpenAI), Gemini (Google), Perplexity, DeepSeek, and GitHub Copilot for all projects. DeepSeek is approved as of May 2026 for the AI-Driven Research SOP (multi-model verification + human audit). Use DeepSeek only via Western-hosted endpoints (Together.ai, Fireworks, OpenRouter, Hyperbolic, or self-hosted) when handling sensitive data.

DEFAULT

Company-Provided

Perplexity Pro + ChatGPT (OpenAI) + Claude (Anthropic)

These are the standard tools for all intern coding and project work. Access provided by your Team Lead.

OPTIONAL

Personal Accounts

  • - ChatGPT Plus / ChatGPT Team (OpenAI)
  • - Claude Pro / Claude Max (Anthropic)
  • - Gemini 3
  • - Other AI assistants you own
๐ŸŽ“

FREE Student Programs

Students can access over $2,000 in free tools through verified student programs.

Perplexity Pro (Students)

Free 1-year Pro access with student verification

Google Gemini (Students)

Free 1-year AI Pro with .edu email

โ˜„๏ธ

Comet Browser Required for Advanced Features

Download the Comet Browser (now FREE on iOS, Android, Windows, Mac) to access the assistant feature and create prompts with GPT-5.5 and Gemini 3.1 Pro built in as tools for creating projects and markdown files.

Comet Browser Features:

  • - GPT-5.5 and Gemini 3.1 Pro included as tools
  • - Create projects and markdown files directly
  • - Automate website browsing
  • - Input data into the browser programmatically
  • - Assistant feature for guided workflows
Download Comet Browser

Training Focus

Our training emphasizes Claude, OpenAI/ChatGPT, and Gemini workflows, PRDs, project context files, and SOPs for comprehensive AI-powered development.

Multi-Platform SOPs Available

This site provides setup guides and SOPs for all three major AI platforms across macOS, Linux, and Windows:

Claude

Opus 4.7 / Sonnet 4.6 / Haiku 4.5

Gemini 3.1 Pro

Google AI Pro

GPT-5.5

OpenAI ChatGPT

DeepSeek V4

Western-host only

Compliance Requirements

No matter which AI tool you use, all outputs must comply with:

  • Security rules: Never include secrets, API keys, or PII in prompts
  • Development SOP: PRD โ†’ project context files โ†’ TDD โ†’ Code Review workflow
  • Handoff standards: Code must be maintainable by any team member

When to Use Each Tool

๐ŸŸ 

Claude

Best for: Code Generation & Refactoring

  • โ€ข Complex code generation
  • โ€ข Large-scale refactoring
  • โ€ข Architecture decisions
  • โ€ข Code review & debugging
  • โ€ข Documentation writing
  • โ€ข Multi-file changes

Best for coding tasks

๐Ÿ”ต

Gemini

Best for: Research & Exploration

  • โ€ข Web search integration
  • โ€ข Documentation research
  • โ€ข API exploration
  • โ€ข Learning new frameworks
  • โ€ข Comparing technologies
  • โ€ข Finding code examples

Great for research tasks

๐ŸŸข

ChatGPT

Best for: Brainstorming & Planning

  • โ€ข Brainstorming ideas
  • โ€ข Project planning
  • โ€ข Quick explanations
  • โ€ข General questions
  • โ€ข Writing assistance
  • โ€ข Creative tasks

Good for ideation

Model Strengths Comparison

Capability๐ŸŸ  Claude๐Ÿ”ต Gemini๐ŸŸข ChatGPT
Code Generationโญโญโญโญโญโญโญ
Web Searchโญโญโญโญโญโญโญ
Context Lengthโญโญโญโญโญโญโญโญ
File Handlingโญโญโญโญโญโญโญ
Multimodal (Images)โญโญโญโญโญโญโญ
IDE Integrationโญโญโญโญโญโญโญโญ

Recommended Multi-Tool Workflow

๐Ÿ”ต

Gemini

Research

โ†’
๐ŸŸข

ChatGPT

Plan

โ†’
๐ŸŸ 

Claude Code

Build

๐Ÿ“ฑ AI Apps for Mobile

Download AI assistants directly to your phone

What phone are you using?

๐Ÿค–

Google Play Store - AI App Links

Tip: All these AI apps are free to download. Some features may require a subscription, but basic functionality is available for free. Students can get Perplexity Pro free for 1 year with student verification, and access 30+ more free student programs!

โš  DeepSeek policy (May 2026): The official DeepSeek mobile apps and chat.deepseek.com route to PRC-hosted endpoints - do not paste customer data, credentials, PII, or company source. For approved company use, route DeepSeek V4 / R1 through Together.ai, Fireworks, OpenRouter, Hyperbolic, or self-hosted endpoints. See the AI-Driven Research SOP for details.

๐Ÿง 

Claude AI Models Guide

Updated May 2026: Opus 4.7, Sonnet 4.6, Haiku 4.5, Extended Thinking, and Deep Research

New - April 16, 2026

Claude Opus 4.7 is now Anthropic's flagship

Replaces Opus 4.6 as flagship. Improved coding (SWE-bench), agent durability, and tool use. Same API price as 4.6 ($5/M input, $25/M output). Model ID: claude-opus-4-7. Sonnet 4.6 and Haiku 4.5 remain unchanged.

Read Announcement โ†’
FLAGSHIP - APR 2026

Claude Opus 4.7

Anthropic's newest flagship (April 16, 2026). Best for complex coding, architecture decisions, and multi-file refactoring. Direct upgrade from Opus 4.6 with no API price change.

Best at:Code generation, debugging, architecture
Context:200K tokens input (1M for Enterprise)
Speed:Slower but more thorough
Tokens:Uses more tokens per response (higher cost)

Use for: Claude Code CLI, agent teams, complex refactoring, code review, multi-step implementations, and any task where quality matters more than speed.

BALANCED

Claude Sonnet 4.6

The ideal balance of speed, quality, and cost. Best for everyday research, writing, and standard development tasks.

Best at:Research, writing, general coding
Context:200K tokens input
Speed:Fast responses
Tokens:More cost-efficient per response

Use for: Extended thinking + deep research, everyday questions, writing tasks, standard code generation, and any task where speed and cost efficiency matter.

TaskOpus 4.7Sonnet 4.6Recommendation
Complex code generationBestGoodUse Opus for multi-file, complex logic
Research & fact-findingOverkillBestSonnet + extended thinking
Architecture decisionsBestGoodUse Opus for critical design work
Quick questions & chatWastefulBestSonnet saves tokens
Deep researchGoodBestSonnet + deep research mode
Agent teams (CLI)Best (Lead)Best (Workers)Mix: Opus lead + Sonnet teammates
Code review / debuggingBestGoodOpus catches more subtle issues
Writing & documentationGoodBestSonnet is fast and high-quality

Cost-Saving Strategy

Opus 4.7 uses significantly more tokens per response than Sonnet 4.6. For most everyday tasks (research, writing, simple code), Sonnet 4.6 with extended thinking delivers comparable quality at a fraction of the token cost. Reserve Opus 4.7 for complex coding tasks, architecture decisions, and agent team leads where the extra reasoning depth justifies the additional token usage.

Quick Reference: Which Model When?

Opus 4.7 for Coding

Complex code, refactoring, debugging, architecture, agent team leads

Sonnet 4.6 for Research

Extended thinking, deep research, writing, general tasks, agent team workers

Mix Models for Teams

Opus lead coordinates, Sonnet teammates execute - best cost/performance ratio

New SOP - May 2026

๐Ÿ”ฌ AI-Driven Research SOP (Multi-Model + Human Audit)

The official Sports Media research workflow. Every research deliverable runs through five stages: Perplexity โ†’ Gemini โ†’ DeepSeek โ†’ Claude โ†’ Human Audit. The point is independent cross-verification - never trust a single model.

Why a multi-model SOP?

  • Single-model bias is real. Each frontier model has different training data, refusal patterns, and blind spots.
  • Independent cross-checking with three reasoning families (Anthropic, Google, DeepSeek/OpenAI) catches hallucinations a single model would not.
  • Cost: pushing the bulk to Sonnet 4.6 + DeepSeek and only using Opus 4.7 / Gemini 3.1 Pro for the synthesis cuts API spend ~60-80%.
  • Audit trail: every step logs sources + the prompt, so any teammate can reproduce or override the conclusion.

Council Pattern (advanced, May 2026)

For high-stakes questions, run Gemini 3.1 Pro and DeepSeek V4-Pro in parallel on the identical Perplexity sources, then use Claude Opus 4.7 as the judge to surface agreement vs divergence. Treat any conflict between the two workers as the verification signal - that is where you spend human audit time. Mirrors the Microsoft Researcher multi-model pattern published in early 2026.

  1. 1

    Frame & Source (Perplexity)

    Perplexity Pro (Sonar / Comet)

    Start with Perplexity because it gives you cited, real-time web answers and named primary sources you can verify yourself. Treat it as a search-and-citation engine, not as the final answer.

    Prompt template: Question: <ONE-SENTENCE QUESTION>. List the 5โ€“10 best primary sources (with publication date) and a 200-word neutral summary. Flag any conflicting claims and cite each one.
    • Required: capture every cited URL into the audit log.
    • Drop any claim Perplexity cannot back with a primary source.
    • Comet Browser (free, May 2026) lets you cross-check the cited pages without leaving the chat.
  2. 2

    Expand & Multimodal (Gemini)

    Gemini 3.1 Pro + Deep Research

    Use Gemini for multimodal evidence (PDFs, slides, screenshots, video transcripts) and Deep Research for long-running, planned reports. Gemini gives breadth and structure that Perplexity skips.

    Prompt template: Using the sources I attach, build a structured comparison table: claim โ†’ evidence โ†’ confidence โ†’ counter-evidence. Include charts where relevant.
    • Attach every Perplexity source. Do not paste in confidential data unless on Workspace tier.
    • Save the Deep Research export (PDF) to the audit log.
    • AI Studio free tier covers ~10โ€“30 RPM - enough for verification work.
  3. 3

    Cheap Cross-Check (DeepSeek)

    DeepSeek V4 (Flash / Pro) or R1 (Western-hosted)

    DeepSeek's open-weight V4 family (Apr 24, 2026 preview) and R1 reasoning model give a third independent reasoning trace at a fraction of the price. We use it specifically as an adversarial verifier - its job is to disagree.

    Prompt template: Adversarial review: read the attached Gemini comparison table. List every claim where you would push back, including specific counter-evidence. Be skeptical and concrete.
    • POLICY: route only via Western-hosted endpoints - Together.ai, Fireworks AI, OpenRouter, Hyperbolic, Deepinfra, or self-hosted vLLM.
    • Do NOT use chat.deepseek.com or .cn endpoints with any customer or company data - data is logged in PRC.
    • Use V4-Flash for cheap bulk checks (~$0.28/M cache-miss input) and V4-Pro or R1 only when reasoning depth matters.
    • Open-weights (MIT license) means you can self-host on a private GPU for the most sensitive prompts.
    • Legacy IDs deepseek-chat / deepseek-reasoner retire July 24, 2026 - migrate to deepseek-v4-flash / -pro now.
  4. 4

    Synthesize & Code (Claude)

    Claude Opus 4.7 + Sonnet 4.6

    Claude is the synthesis and writing engine. Opus 4.7 (Apr 16, 2026 release) does the final reasoning and any code; Sonnet 4.6 handles the bulk drafting because it is faster and cheaper. Claude Code CLI runs the workflow end-to-end with version control.

    Prompt template: Synthesize the Perplexity sources, the Gemini comparison table, and the DeepSeek adversarial review into a final memo. Cite each source. Mark every claim as Confirmed / Disputed / Unverifiable. Output as Markdown with a sources section.
    • Opus 4.7 for the final pass; Sonnet 4.6 for everything earlier.
    • Run via Claude Code CLI inside the project repo so the memo lands in /docs and is version-controlled.
    • Use Extended Thinking when the question is ambiguous.
  5. 5

    Human Audit (Required)

    You (the human)

    No AI output ships without a human audit. The audit is what turns AI research into a Sports Media deliverable. Skipping this step is a policy violation.

    Prompt template: For every Confirmed claim, open at least one cited source and verify the wording. For every Disputed claim, decide and document the resolution. Sign off in the audit log with name, date, and version hash.
    • Verify at least the top 5 highest-impact citations against the primary source.
    • Re-run Step 3 (DeepSeek adversarial) if you spot a gap - cheap insurance.
    • PII / customer data / credentials must NEVER leave the corporate Passbolt vault during research.
    • Commit the final memo and audit log to the repo. PR review is mandatory.

โš ๏ธ DeepSeek Hosting Policy (Required Reading)

DeepSeek is approved for the AI-Driven Research SOP as of May 2026, but only via Western-hosted endpoints. The policy is hosting-based, not model-based: the open-weights are fine; the .cn API is not.

  • Approved hosts: Together.ai, Fireworks AI, OpenRouter, Hyperbolic, Deepinfra, or self-hosted vLLM/TGI.
  • Not approved for company data: chat.deepseek.com, the official .cn API, the iOS / Android DeepSeek apps for Sports Media work.
  • For DoW / DoD work: defer to your contracting officer. Pentagon-contracted vendors (OpenAI, Anthropic, Google, xAI) remain the default for CUI/classified workflows.

๐Ÿงพ Audit Log Template (commit to /docs/research-audit/)

# AI Research Audit - <slug>
date: 2026-05-07
author: <name>
question: <one sentence>

## 1. Perplexity sources
- <url> | <title> | <date> | <pull quote>

## 2. Gemini comparison
- attached: gemini-research-<slug>.pdf

## 3. DeepSeek adversarial review
- host: together.ai (model: deepseek-ai/DeepSeek-V3)
- counter-claims: ...

## 4. Claude synthesis
- model: claude-opus-4-7
- output: docs/research/<slug>.md (commit <sha>)

## 5. Human audit
- verified citations: 5 / 5
- disputed claims resolved: <details>
- signed: <name>, 2026-05-07

AI Development Deep Dive

Project methodology, templates, prompts, and advanced workflows

๐Ÿง 

Claude Code Project Methodology

The PSB System: Plan โ†’ Setup โ†’ Build

Before Opening Claude Code

Question 1: What is the project goal?

  • โ€ข Learning a new technology
  • โ€ข Validating an idea with customers
  • โ€ข Building an alpha product/feature
  • โ€ข Prototyping to test feasibility

Determines what to build vs. skip. Prototypes skip auth, databases, styling.

Question 2: What are the milestones?

  • โ€ข Define what version 1 looks like
  • โ€ข Identify what to defer to v2, v3
  • โ€ข Think MVP first, expand later

Clear scope prevents feature creep and keeps focus.

Pro Tip: Let Claude Help You Plan

Use this prompt to kickstart your planning:

I want to build [project]. What are the three most important
questions I need to answer to build an MVP successfully?

Phase 2: Setup Checklist

  1. 1
    GitHub Repo: Full change history, branch-based development, deploy previews
  2. 2
    Environment Variables: Create .env with all API keys before starting
  3. 3
    CLAUDE.md: Project memory file (see template below)
  4. 4
    Documentation: architecture.md, decisions.md, project-status.md
  5. 5
    MCP Servers: Match to your tech stack (Supabase MCP, etc.)
  6. 6
    Slash Commands: Set up /commit, /pr, custom commands
  7. 7
    Infrastructure: Provision databases, hosting, API keys upfront

Phase 3: Build Workflows

Workflow 1: General

Single feature development

  1. 1. Research
  2. 2. Plan
  3. 3. Implement
  4. 4. Test

Workflow 2: Issue-Based

Split work into GitHub issues

  • โ€ข Organized tracking
  • โ€ข Clear scope per task
  • โ€ข Easy parallel work

Workflow 3: Agent Teams

Opus 4.6 parallel agents

  • โ€ข tmux split panes
  • โ€ข Parallel execution
  • โ€ข Shared task list
Setup guide โ†’
๐Ÿ”

Codebase Architecture Review

Codebase

Get a comprehensive overview of the codebase structure

Review this codebase and explain the architecture. Include:
1. Main directories and their purposes
2. Key files and their roles
3. Data flow patterns
4. Dependencies and external services
5. Any architectural patterns used (MVC, Clean Architecture, etc.)
๐Ÿ”

Find Entry Points

Codebase

Identify where the application starts and routes

Identify all entry points in this application:
1. Main application entry (index, main, app files)
2. API route handlers
3. Event listeners or hooks
4. Scheduled jobs or cron tasks
5. External service integrations
๐Ÿ”

Dependency Analysis

Codebase

Understand package dependencies and their usage

Analyze the dependencies in this project:
1. List all production dependencies and what they're used for
2. Identify any outdated or deprecated packages
3. Flag potential security concerns
4. Suggest any unnecessary dependencies that could be removed
๐Ÿ’ป

Create GitHub Issues from Spec

Development

Convert project specs into actionable GitHub issues

Based on the project specification, create GitHub issues for each feature/task. For each issue include:
- Clear, actionable title
- Detailed description with acceptance criteria
- Labels (feature, bug, enhancement, etc.)
- Estimated complexity (S/M/L)
- Dependencies on other issues

Format as markdown that can be copy-pasted into GitHub.
๐Ÿ’ป

Implement Feature

Development

Step-by-step feature implementation

Variables to replace:

[FEATURE_NAME]
I need to implement [FEATURE_NAME]. Please:
1. First, analyze the existing codebase to understand current patterns
2. Propose an implementation plan with specific files to create/modify
3. Wait for my approval before proceeding
4. Implement following existing code conventions
5. Add appropriate error handling
6. Include any necessary tests
๐Ÿ’ป

Code Review Request

Development

Get a thorough code review with actionable feedback

Review this code for:
1. Logic errors or bugs
2. Security vulnerabilities
3. Performance issues
4. Code style and consistency
5. Missing error handling
6. Opportunities for improvement

Provide specific, actionable feedback with code examples where helpful.
๐Ÿ’ป

Refactor Following Pattern

Development

Refactor code to follow a specific pattern

Variables to replace:

[FILE_OR_COMPONENT][PATTERN_NAME]
Refactor [FILE_OR_COMPONENT] following the [PATTERN_NAME] pattern.
1. Explain the current structure and its issues
2. Show the proposed refactored structure
3. Implement the refactoring step by step
4. Ensure no functionality is broken
5. Update any affected imports or dependencies
๐Ÿงช

Write Tests for Component

Testing

Generate comprehensive tests for a component

Variables to replace:

[COMPONENT_NAME]
Write tests for [COMPONENT_NAME]. Include:
1. Unit tests for all public methods/functions
2. Edge cases and error conditions
3. Integration tests if applicable
4. Mock external dependencies appropriately
5. Follow existing test patterns in the codebase

Use the testing framework already configured in the project.
๐Ÿงช

Test Coverage Analysis

Testing

Identify gaps in test coverage

Analyze the test coverage of this codebase:
1. Which files/functions lack tests?
2. What edge cases are missing?
3. Are there any critical paths without coverage?
4. Prioritize which tests to write first based on risk
๐Ÿงช

Fix Failing Test

Testing

Debug and fix a failing test

Variables to replace:

[TEST_NAME][ERROR_MESSAGE]
This test is failing: [TEST_NAME]

Error message: [ERROR_MESSAGE]

Please:
1. Analyze why the test is failing
2. Determine if the test or the code is wrong
3. Fix the appropriate part
4. Ensure no other tests are affected
๐Ÿ›

Debug Error Message

Debugging

Systematic error debugging

Variables to replace:

[ERROR_MESSAGE][WHERE_IT_OCCURS]
Debug this error: [ERROR_MESSAGE]

Context: [WHERE_IT_OCCURS]

Please:
1. Identify the likely cause
2. Trace the error through the code
3. Propose a fix with explanation
4. Suggest how to prevent similar errors
๐Ÿ›

Performance Investigation

Debugging

Find and fix performance bottlenecks

Variables to replace:

[FEATURE/PAGE]
This [FEATURE/PAGE] is slow. Please:
1. Identify potential performance bottlenecks
2. Check for unnecessary re-renders (React) or queries
3. Look for N+1 query problems
4. Suggest optimizations with priority order
5. Implement the top 3 fixes
๐Ÿ›

Memory Leak Investigation

Debugging

Find memory leaks in the application

Investigate potential memory leaks in this codebase:
1. Check for uncleared event listeners
2. Look for unclosed connections or streams
3. Identify circular references
4. Check for improper cleanup in React useEffect
5. Suggest fixes for any issues found
๐Ÿ“š

Update CLAUDE.md

Documentation

Update project memory based on recent changes

Based on the recent changes in this session, update CLAUDE.md:
1. Add any new commands or scripts
2. Document new architectural decisions
3. Update the "Current Focus" section
4. Add any new gotchas discovered
5. Remove outdated information
๐Ÿ“š

Generate API Documentation

Documentation

Create comprehensive API docs

Variables to replace:

[ENDPOINT_OR_FILE]
Generate API documentation for [ENDPOINT_OR_FILE]:
1. Endpoint URL and method
2. Request parameters and body schema
3. Response format with examples
4. Error codes and messages
5. Authentication requirements
6. Usage examples with curl/fetch
๐Ÿ“š

Write Technical Decision Doc

Documentation

Document a technical decision with rationale

Variables to replace:

[DECISION_MADE]
Document this technical decision for decisions.md:

Decision: [DECISION_MADE]

Include:
1. Context: What problem were we solving?
2. Options Considered: What alternatives did we evaluate?
3. Decision: What did we choose and why?
4. Consequences: What are the tradeoffs?
5. Date and author
๐Ÿ”€

Generate Commit Message

Git & PRs

Create a descriptive commit message

Based on the staged changes, generate a commit message following this format:
- First line: type(scope): brief description (50 chars max)
- Blank line
- Body: detailed explanation of what and why (wrap at 72 chars)

Types: feat, fix, docs, style, refactor, test, chore
๐Ÿ”€

PR Description

Git & PRs

Generate a comprehensive PR description

Variables to replace:

[ISSUE_NUMBER]
Based on all commits in this branch, generate a PR description:
## Summary
[Brief overview of changes]

## Changes
- [List of specific changes]

## Testing
- [ ] Unit tests added/updated
- [ ] Manual testing performed

## Screenshots (if applicable)

## Related Issues
Closes #[ISSUE_NUMBER]
๐Ÿ”€

Resolve Merge Conflict

Git & PRs

Help resolve git merge conflicts

Variables to replace:

[FILE_NAME]
I have merge conflicts in [FILE_NAME]. Please:
1. Show me both versions of the conflicting code
2. Explain what each version is trying to do
3. Suggest the correct resolution
4. Explain any potential side effects of the merge

Tips for Using Prompts

  • โ€ข Replace bracketed variables [LIKE_THIS] with your specific values
  • โ€ข Combine prompts for complex tasks (e.g., Review + Refactor)
  • โ€ข Add context about your project for better results
  • โ€ข Use the # command to save learnings from Claude's responses to CLAUDE.md

What is MCP (Model Context Protocol)?

MCP gives Claude Code access to external tools and data sources. Think of it as plugins that let Claude interact with databases, APIs, file systems, and more directly. Latest spec: 2025-11-25 (OAuth 2.1 resource-server model). MCP was donated to the Linux Foundation Agentic AI Foundation in December 2025 and is now supported by Claude Code, Claude Desktop, ChatGPT (Developer Mode), Cursor, Windsurf, Zed, VS Code, and Cline.

๐Ÿ—„๏ธ

Database Access

Query and modify databases directly

๐Ÿ”„

Schema Updates

Automatic schema changes during iteration

โšก

Real-time Tools

Integrate with live services

โš ๏ธ MCP Security - Read Before Installing Servers

February 2026 saw 8,000+ exposed MCP servers in the wild. Treat third-party MCP servers like any untrusted code. Best practices:

  • Prefer first-party (vendor-published) servers over community wrappers.
  • Scope tokens narrowly. Use OAuth 2.1 + PKCE for remote/HTTP MCPs.
  • Review the tool descriptions - prompt injection via tool descriptions is a real attack.
  • Use --scope project for committed configs and --scope user for personal/global.
  • See the OWASP MCP Security Cheat Sheet and Anthropic's MCP security best practices.
๐ŸŸข

Supabase MCP Setup

RECOMMENDED

Direct Supabase access (auth, database, storage, edge functions) via Supabase's first-party MCP server. Enables automatic schema updates and queries during rapid iteration.

All platforms (Claude Code v2.1+):

claude mcp add supabase --scope user --transport stdio -- \
  npx -y @supabase/mcp-server-supabase --access-token=YOUR_TOKEN

Token: Generate a personal access token in your Supabase Dashboard โ†’ Account โ†’ Access Tokens. Use --scope project instead of userto commit the config to .mcp.json for the team.

๐Ÿ”ฅ Top First-Party MCP Servers (April 2026)

๐Ÿ›ก๏ธ Cybersecurity MCP Servers

๐Ÿ“

Filesystem MCP (Reference)

Extended file operations - first-party reference server.

claude mcp add filesystem --scope user -- \
  npx -y @modelcontextprotocol/server-filesystem /path/to/project
๐ŸŒ

Fetch MCP (Reference)

Fetch and convert URL content for the model.

claude mcp add fetch --scope user -- \
  npx -y @modelcontextprotocol/server-fetch
๐Ÿ™

GitHub MCP (Official)

Repos, issues, PRs - via GitHub's first-party server.

claude mcp add github --transport http \
  --scope user https://api.githubcopilot.com/mcp/
๐ŸŽญ

Playwright MCP (Microsoft)

Browser automation for tests, scraping, agentic flows.

claude mcp add playwright --scope user -- \
  npx -y @playwright/mcp@latest

Managing MCPs (Claude Code v2.1+)

claude mcp list- List installed MCPs
claude mcp get <name>- Inspect a server's config
claude mcp remove <name>- Remove an MCP
--scope local | project | user- Scope of the install (local default; project commits .mcp.json)
--transport stdio | http | sse- Transport (stdio for local, http for remote with OAuth 2.1)
mcp__alwaysLoad- Settings flag to force-load servers at session start

๐Ÿงฉ Skills & Plugins (Claude Code, 2026)

Beyond MCP, Claude Code now ships Skills (auto-invoked instruction bundles in~/.claude/skills/) and the/plugin system, which can ship Skills + Slash Commands + Hooks + MCP server configs as a single installable bundle from a marketplace (/plugin marketplace add owner/repo). Theofficial Anthropic marketplacehas 100+ plugins.

Rule of thumb: MCP = external tools (network/process). Skills = context-aware instruction bundles, model-invoked. Slash commands = explicit user-typed prompts.Plugins = distribution wrapper that combines all three.

These documents are separate from CLAUDE.md. AI tools read them when needed for context. Keep them updated as the project evolves.

๐Ÿ—๏ธ
docs/architecture.md

System design, component relationships, and data flow

# Architecture Overview

## System Design

### High-Level Architecture
```
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚                        Frontend                              โ”‚
โ”‚  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”         โ”‚
โ”‚  โ”‚   Next.js   โ”‚  โ”‚   React     โ”‚  โ”‚  Tailwind   โ”‚         โ”‚
โ”‚  โ”‚  App Router โ”‚  โ”‚ Components  โ”‚  โ”‚    CSS      โ”‚         โ”‚
โ”‚  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜         โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                            โ”‚
                            โ–ผ
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚                      API Layer                               โ”‚
โ”‚  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”         โ”‚
โ”‚  โ”‚  API Routes โ”‚  โ”‚  Supabase   โ”‚  โ”‚    Edge     โ”‚         โ”‚
โ”‚  โ”‚  /api/*     โ”‚  โ”‚   Client    โ”‚  โ”‚  Functions  โ”‚         โ”‚
โ”‚  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜         โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
                            โ”‚
                            โ–ผ
โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
โ”‚                     Supabase                                 โ”‚
โ”‚  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”  โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”         โ”‚
โ”‚  โ”‚ PostgreSQL  โ”‚  โ”‚    Auth     โ”‚  โ”‚   Storage   โ”‚         โ”‚
โ”‚  โ”‚  + RLS      โ”‚  โ”‚             โ”‚  โ”‚             โ”‚         โ”‚
โ”‚  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜  โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜         โ”‚
โ””โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”˜
```

## Component Architecture

### Frontend Components
| Component | Purpose | Location |
|-----------|---------|----------|
| Layout | Root layout with nav/footer | `app/layout.tsx` |
| [Component] | [Purpose] | `components/[name].tsx` |

### Data Flow
1. User interacts with React component
2. Component calls API route or Supabase client
3. Supabase processes request with RLS policies
4. Data returned and cached appropriately

## Database Schema

### Tables
```sql
-- Core tables
[Add your table definitions]

-- Relationships
[Add foreign key relationships]
```

### Row Level Security
| Table | Policy | Description |
|-------|--------|-------------|
| [table] | [policy] | [description] |

## Authentication Flow
1. User signs in via Supabase Auth
2. Session stored in secure cookie
3. Middleware validates session on protected routes
4. RLS policies enforce data access

## External Services
| Service | Purpose | Integration Point |
|---------|---------|------------------|
| Vercel | Hosting | Automatic deployment |
| Supabase | Backend | Database, Auth, Storage |

## Security Considerations
- All tables have RLS enabled
- Service role key used only in Edge Functions
- CORS configured for production domains
- Input validation on all endpoints

## Performance Optimizations
- Server Components for static content
- Dynamic imports for heavy components
- Image optimization via next/image
- Database indexes on frequently queried columns

## Deployment Architecture
```
GitHub Repo
    โ”‚
    โ”œโ”€โ”€ Push to main โ”€โ”€โ–บ Vercel (Production)
    โ”‚
    โ””โ”€โ”€ Push to dev โ”€โ”€โ”€โ–บ Vercel (Preview)
```

Recommended Project Structure

project/
โ”œโ”€โ”€ CLAUDE.md                    # Project memory (always in context)
โ”œโ”€โ”€ docs/
โ”‚   โ”œโ”€โ”€ architecture.md          # System design
โ”‚   โ”œโ”€โ”€ decisions.md             # Technical decision log
โ”‚   โ”œโ”€โ”€ project-status.md        # Progress and handoff notes
โ”‚   โ””โ”€โ”€ features/
โ”‚       โ”œโ”€โ”€ auth.md              # Authentication feature
โ”‚       โ”œโ”€โ”€ dashboard.md         # Dashboard feature
โ”‚       โ””โ”€โ”€ [feature-name].md    # Other features
โ”œโ”€โ”€ supabase/
โ”‚   โ”œโ”€โ”€ migrations/              # Database migrations
โ”‚   โ””โ”€โ”€ functions/               # Edge Functions
โ””โ”€โ”€ src/                         # Source code

Keep Current

Update project-status.md at the end of each session

Document Decisions

Log technical decisions as you make them

Reference in CLAUDE.md

Link to docs from CLAUDE.md, don't duplicate

Select AI Platform to Set Up:

AI Developer Setup for Windows

This guide covers setup for all three AI platforms:

Claude
Gemini 3
ChatGPT (GPT-5.5)

Quickstart Checklist

  • [ ]Install Node.js LTS from nodejs.org, verify with node -v / npm -v
  • [ ]Install Claude Code CLI: npm install -g @anthropic-ai/claude-code
  • [ ]Run claude login and test with a hello world prompt
  • [ ]Install Claude Code extension in VS Code (run claude in integrated terminal)
  • [ ]Install Supabase CLI using Scoop, verify with supabase --version
  • [ ]Install Docker Desktop with WSL2, verify with docker run hello-world
  • [ ]Install Git, configure GitHub identity, validate with git --version
  • [ ]Install Vercel CLI: npm i -g vercel, login with vercel login
  • [ ]Install GitHub CLI: winget install GitHub.cli, authenticate with gh auth login
  • [ ]Login to Supabase: supabase login

Node.js & Claude CLI Installation

Step 1: Install Node.js LTS

Download and install from nodejs.org (LTS version recommended)

# Verify installation node -v npm -v

Step 2: Install Claude Code CLI

# Install Claude Code globally (no sudo needed on Windows) npm install -g @anthropic-ai/claude-code # Update to latest version claude update # Authenticate with your account claude login # Test with a simple prompt claude "Hello, Claude!"

If you get permission errors, see the npm permissions fix below.

Fix npm Permissions (if needed)

# Create npm global directory mkdir "%USERPROFILE%\.npm-global" # Configure npm to use it npm config set prefix "%USERPROFILE%\.npm-global" # Add to PATH (System Environment Variables): # %USERPROFILE%\.npm-global

Claude in VS Code

Auto-Install via Terminal

Running claude in the VS Code integrated terminal automatically installs the Claude Code extension.

# Open VS Code integrated terminal (Ctrl+`) # Run claude to auto-install extension claude

Manual Install (Alternative)

  1. Open VS Code Extensions (Ctrl+Shift+X)
  2. Search for "Claude Code"
  3. Click Install

Usage Patterns

  • Right-click menu: Select code โ†’ Right-click โ†’ "Ask Claude"
  • Command Palette: Ctrl+Shift+P โ†’ Type "Claude Code" for available commands
  • Side Panel: Use "Claude Code: Open in Side Bar" to pin Claude for autonomous work

Supabase CLI (Windows via Scoop)

# Run PowerShell as Administrator # Set execution policy Set-ExecutionPolicy RemoteSigned -Scope CurrentUser # Install Scoop irm get.scoop.sh | iex # Add Supabase bucket scoop bucket add supabase https://github.com/supabase/scoop-bucket # Install Supabase CLI scoop install supabase # Verify installation supabase --version # Initialize a new Supabase project supabase init # Start local Supabase stack supabase start # Stop when done supabase stop # Login to Supabase Cloud supabase login

Tip: Run supabase start in your project directory to spin up a local Postgres database, Auth, and Storage services. Run supabase login to link with your Supabase Cloud account.

Docker Desktop + WSL2

Install WSL2

# Run in PowerShell as Administrator wsl --install # Restart your computer after installation

Install Docker Desktop

  1. Download Docker Desktop from docker.com
  2. Run the installer and enable WSL2 integration when prompted
  3. Restart your computer if required
  4. Launch Docker Desktop

Verify Docker & Run n8n

# Test Docker installation docker run hello-world # Check Docker version docker --version docker-compose --version # Run n8n workflow automation docker run -it --rm \ --name n8n \ -p 5678:5678 \ -v ~/.n8n:/home/node/.n8n \ n8nio/n8n

Access n8n at: http://localhost:5678

Git & GitHub Basics

Install & Configure Git

# Download Git from https://git-scm.com/download/win # Run installer with default options # Configure your identity git config --global user.name "Your Name" git config --global user.email "your.email@example.com" # Verify installation git --version

Basic Git Commands

# Clone a repository git clone https://github.com/username/repo.git # Check status git status # Stage and commit changes git add . git commit -m "Your commit message" # Push to remote git push origin main

Vercel CLI Setup

Install & Authenticate

# Install Vercel CLI globally npm i -g vercel # Login to Vercel vercel login # Verify installation vercel --version

Deploy Commands

# Deploy preview (creates unique URL) vercel # Deploy to production vercel --prod # Link existing project vercel link # Pull environment variables vercel env pull .env.local

GitHub CLI Setup

Install & Authenticate

# Install via winget (recommended) winget install GitHub.cli # Or via Scoop scoop install gh # Authenticate with GitHub gh auth login # Verify installation gh --version

Common Commands

# Clone a repository gh repo clone owner/repo # Create a new repository gh repo create my-project --public # Create a pull request gh pr create --title "Feature: Add login" --body "Description" # View open issues gh issue list # Create an issue gh issue create --title "Bug: Fix auth" --body "Details"

Tip: The GitHub CLI integrates perfectly with Claude Code. Use gh commands to manage issues and PRs directly from your terminal.

Advanced Tips

Project Launch Sequence (Vite/Next.js)

# Example path - replace with your own repo path cd C:\Users\YourUsername\GitHub\your-project\client # Install dependencies npm install # Start development server npm run dev

Note: Replace the path above with your actual project location.

Autonomous Mode (Advanced Users Only)

Warning: This command gives Claude extensive permissions. Use with caution and only when you understand the implications.

claude --dangerously-skip-permissions

Crash Recovery

# Kill stuck Node processes taskkill /f /im node.exe # Remove Next.js cache rmdir /s /q .next # Reinstall and restart npm install npm run dev

Troubleshooting

Permission Errors (EACCES)

Fix npm permissions without sudo:

npm config set prefix "%USERPROFILE%\.npm-global" # Add %USERPROFILE%\.npm-global to PATH

WSL Connection Issues

Restart WSL service:

wsl --shutdown && wsl

Claude Authentication Failed

Clear credentials and re-authenticate:

rmdir /s /q %USERPROFILE%\.claude claude login

Node Version Conflicts

Use nvm-windows to manage versions:

# Install nvm-windows from github.com/coreybutler/nvm-windows nvm install 20 nvm use 20

Gemini 3 CLI Setup

Install Gemini CLI

# Install Gemini CLI globally via npm npm install -g @google/gemini-cli # Authenticate with your Google account gemini auth login # Verify installation gemini --version # Test with a prompt gemini "Hello, Gemini!"

Gemini CLI Usage

# Start interactive chat gemini chat # Analyze an image gemini "Describe this image" --image ./screenshot.png # Generate code gemini "Write a Python function to sort a list" # Use in a project directory cd C:\Users\YourUsername\your-project gemini "Explain this codebase"

Student Offer: Get Gemini 3.1 Pro free for 1 year with the Google AI Pro student plan. See the Gemini Student Plan section above for signup details.

ChatGPT CLI Setup

Install OpenAI CLI

# Install OpenAI CLI via pip (requires Python) pip install openai # Set your API key (get from platform.openai.com) # In PowerShell: $env:OPENAI_API_KEY="your-api-key-here" # Or add to System Environment Variables permanently: # Settings > System > About > Advanced system settings # Environment Variables > New > OPENAI_API_KEY

ChatGPT CLI Usage

# Use the openai command openai api chat.completions.create -m gpt-4o -g user "Hello, ChatGPT!" # For interactive use, consider chatgpt-cli wrapper npm install -g chatgpt-cli chatgpt "Write a function to parse JSON"

Tip: Use the Comet Browser (now FREE on iOS/Android/Windows/Mac) for integrated access to GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro without managing separate CLI tools.

Final Validation Checklist

Run these commands to verify your setup is complete:

# Claude claude --version # Gemini gemini --version # General tools node -v npm -v git --version docker run hello-world supabase --version

All commands should execute without errors. If any fail, refer to the troubleshooting section above.

Authentication Failed

Login token expired or invalid

# Clear existing auth and re-login supabase logout supabase login # Verify auth status supabase projects list

Migration Conflicts

Local and remote schemas out of sync

# Pull remote schema changes supabase db pull # Reset local database to match migrations supabase db reset # If conflicts persist, check migration history supabase migration list

Warning: db reset will delete all local data. Export important data first.

Local Dev Not Starting

Docker containers failing to start

# Stop all Supabase containers supabase stop --no-backup # Check Docker is running docker ps # Start fresh supabase start # If port conflicts: # Check what's using ports 54321-54329 lsof -i :54321 # macOS/Linux netstat -ano | findstr :54321 # Windows

Edge Function Deploy Failed

Function not deploying to cloud

# Check function syntax locally supabase functions serve function-name # Deploy with verbose output supabase functions deploy function-name --debug # Check function logs supabase functions logs function-name

Type Generation Failing

Cannot generate TypeScript types

# Ensure local Supabase is running supabase start # Generate types (local) supabase gen types typescript --local > src/types/supabase.ts # Generate types (remote) supabase gen types typescript --project-id YOUR_PROJECT_ID > src/types/supabase.ts

Claude + n8n Engineering Workflow (SOP Summary)

1PRD Requirements

  • Every production project must have a PRD based on the canonical template
  • PRD must define MVP scope, user stories, acceptance criteria, and success metrics
  • PRD must be approved before Claude/n8n work begins

2Plan Mode Workflow

  1. Claude proposes an implementation plan
  2. Human reviews and approves the plan
  3. Claude writes code
  4. Tests are written (TDD approach) and run
  5. Human code review required before merge (security, performance, edge cases)

Minimum 80% test coverage required for critical paths

3n8n Usage Guidelines

  • Use n8n to orchestrate Claude API calls where appropriate
  • Implement error handling and retry logic in workflows
  • Enable prompt logging for debugging and audit trails
  • Monitor workflows for errors, latency spikes, and anomalies

!Security & Privacy

  • Never include secrets or PII in prompts
  • Keep API keys and credentials in environment variables or secrets managers
  • Use separate keys for dev/staging/production environments
  • Rotate credentials regularly and revoke unused keys

Intern Workflow & Communication SOP

Slack Communication

Use Slack for all team coordination. Key channels for interns:

#computer-science#5-training-new-interns

Requesting Access (AI Tools, Repos, Supabase, Env Vars)

When requesting access to any resource, include the following in your message:

  • Title/Purpose:What you need access to and why
  • Owner:Who is responsible for granting access
  • Audience:Which channel/roles should be notified
  • Due Date:When you need it by (use MST timezone)
  • Attachments:Links to relevant Drive docs or GitHub repos

Communication Best Practices

  • Keep prompts and messages concise and actionable
  • Never paste API keys or secrets into Slack messages
  • Use threads for follow-up discussions to keep channels organized
  • Use reactions for quick status updates:
    • - Acknowledged / seen
    • - Done / completed
    • - Blocked / need help

Quick Reference

When in doubt, ask in #5-training-new-interns or reach out to your Team Lead directly. It's better to ask questions than to make assumptions that could impact the project.