The most complete Claude Code toolkit in the wild. Battle-tested behavioral rules, 63 slash commands, 327 skills, 34 custom agents, stage-gate pipelines, multi-model orchestration, and reusable thinking frameworks -- all derived from 2,990 sessions and 3,307 commits across 5 months of intensive AI-augmented development.
3,307 commits taught us how AI agents fail. This is the operating system that prevents it.
Squire is a collection of rules, patterns, templates, and tools that make AI coding agents (Claude Code, Cursor, Copilot, etc.) more reliable, more productive, and less likely to waste your time.
It's not a product. It's an operating system -- a set of files you drop into your project or your global config that change how your agent behaves.
What you get:
| Component | Count | What It Does |
|---|---|---|
squire.md |
1 | The flagship. A complete behavioral ruleset for AI agents -- drop it into your project root or ~/.claude/CLAUDE.md |
| Commands | 63 | Claude Code slash commands: /ship, /deploy, /research, /visualize, /blueprint, /fix, /test, and 56 more |
| Skills | 327 | Specialized skills across engineering, marketing, finance, AI/ML, design, and operations |
| Agents | 34 | Custom agents with tool access for architecture, security, DevOps, and more |
| Pipeline | 11 stages | Stage-gate build system with gate questions, agent-native additions, and branch hygiene |
| Patterns | 7 | Battle-tested behavioral rules and multi-model orchestration (Director/Builder) |
| Prompts | 6 | Thinking frameworks for code review, debugging, security, performance, testing, and ship readiness |
| Templates | 3 | The Triad: VISION.md + SPEC.md + BUILDING.md -- self-correcting product documentation |
| Doc Templates | 6 | Feature specs, ADRs, deployment checklists, implementation plans |
| Workspace | 6 generators | File-based project organization with generator scripts for tasks, bugs, features, and ADRs |
| BUILDING-SETUP.md | 1 | Self-installing build journal -- drop into any project, it sets itself up and maintains itself |
Copy squire.md to your project root. That's it. Claude Code reads it automatically.
curl -fsSL https://raw.githubusercontent.com/eddiebelaval/squire/main/squire.md > squire.mdOr for global rules (all projects):
curl -fsSL https://raw.githubusercontent.com/eddiebelaval/squire/main/squire.md >> ~/.claude/CLAUDE.mdClone the repo and run the installer:
git clone https://github.com/eddiebelaval/squire.git
cd squire
chmod +x install.sh
./install.shEverything is standalone. Copy individual files:
# Just the thinking frameworks
cp squire/prompts/*.md ~/.claude/prompts/
# Just the slash commands
cp squire/commands/*.md ~/.claude/commands/
# Just the build journal
cp squire/BUILDING-SETUP.md ./
# Just the workspace generators
cp -r squire/workspace/generators/ ./generators/These are the behavioral corrections that emerged from analyzing real sessions. Each one addresses a specific failure pattern where AI agents consistently waste time or introduce bugs.
| # | Rule | Failure It Prevents |
|---|---|---|
| 1 | Default to implementation | Agent plans endlessly instead of building |
| 2 | Plan means plan | User asks for a plan, gets an audit or exploration instead |
| 3 | Preflight before push | Broken code pushed to remote without verification |
| 4 | Investigate bugs directly | Agent dismisses errors as "stale cache" without looking |
| 5 | Scope changes to the target | Config change for one project applied globally |
| 6 | Verify after each edit | Batch edits create cascading type errors |
| 7 | Visual output verification | Agent re-reads CSS instead of checking rendered output |
| 8 | Check your environment | CLI command runs against wrong project/environment |
| 9 | Don't over-engineer | Simple feature gets unnecessary abstractions |
Full descriptions with friction patterns: patterns/behavioral-rules.md
A three-document system that replaces your dead PRD. Three documents that stay in sync because they reference each other. Any two can reconstruct the third.
VISION.md
(FUTURE)
/ \
/ \
/ THE \
/ WORK \
/ \
SPEC.md -------- BUILDING.md
(PRESENT) (PAST)
- VISION.md -- Where it's going. Soul, pillars (REALIZED/PARTIAL/UNREALIZED), anti-vision, edges.
- SPEC.md -- Where it is right now. Testable contract with drift detection (CURRENT/DRIFTED/STALE).
- BUILDING.md -- How it got here. Self-installing, auto-updating build journal with check-in interviews.
The gap between VISION and SPEC IS the roadmap. No separate roadmap document. The unrealized pillars are what you build next. The delta between "what we want" and "what we have" tells you everything.
Self-correcting: When one document drifts, the others expose it. SPEC says "supports real-time" but VISION never mentioned it? Either VISION needs updating or the feature is scope creep.
Self-installing: Drop BUILDING-SETUP.md into any project root. It explores your codebase, interviews you about the origin story, generates a personalized BUILDING.md, then deletes itself.
Use /reconcile to maintain the Triad conversationally -- it detects drift, interviews you about what changed, and updates all three documents.
Full guide: templates/TRIAD.md
11 stages from concept to production, each with a gate question that must be answered before advancing:
- Concept Lock -- "What's the one-liner?"
- Scope Fence -- "What are we NOT building?"
- Architecture Sketch -- "Draw me the boxes and arrows."
- Foundation Pour -- "Can we deploy an empty shell?"
- Feature Blocks -- "Does this feature work completely, right now?"
- Integration Pass -- "Do all the pieces talk to each other?"
- Test Coverage -- "Are all tests green?"
- Polish & Harden -- "What breaks if I do something stupid?"
- Launch Prep -- "Could a stranger use this?"
- Ship -- "Is it live?"
- Listen & Iterate -- "What did we learn?"
Each stage includes agent-native additions for AI-assisted development. Full spec: pipeline/pipeline.md
A pattern for using two AI models together:
- Director (reasoning model): Plans, reviews, integrates, ships
- Builder (code generation model): Executes scoped implementation tasks
This creates error diversity, context efficiency, and built-in code review. The Director never ships code it hasn't verified.
Failure threshold: 2 Builder failures on the same task = Director takes over.
Full spec: patterns/director-builder.md
Reusable prompt templates for common tasks:
| Framework | When to Use |
|---|---|
code-review.md |
Before approving any PR -- 7-dimension systematic review |
debug-systematic.md |
When encountering a bug -- reproduce, isolate, hypothesize, fix, prevent |
ship-readiness.md |
Before deploying -- 8-section go/no-go checklist |
performance-audit.md |
When something is slow -- measure first, optimize second |
security-audit.md |
Before shipping auth or data features -- STRIDE analysis |
test-strategy.md |
When planning test coverage -- pyramid balance, edge cases, CI integration |
| Command | What It Does |
|---|---|
/ship |
Full delivery pipeline: preflight checks, commit, push, PR, merge |
/fix |
Systematic bug diagnosis and root-cause debugging |
/test |
Browser-based feature testing with Playwright |
/test-verify |
Auto-detect project type and run appropriate tests |
/deploy |
Production deployment with preflight, smoke tests, and rollback |
/deploy-watch |
Monitor deployment status until live |
/start |
Begin new feature with branch creation and planning |
/release |
Promote dev to main with preflight checks and merge verification |
/rollback |
Safe undo of recent changes with soft/hard options |
/cleanup |
Code cleanup workflow (dead code, organization, security) |
/preview |
Pre-commit review with quality checks and risk assessment |
/mobile-check |
Comprehensive mobile viewport testing |
| Command | What It Does |
|---|---|
/visualize |
Interactive HTML visualizations of architecture, workflows, codebases |
/blueprint |
Persistent build plans with progress tracking and parallel batches |
/codebase-map |
Interactive architecture visualization with protocol completion tracking |
/integration-audit |
Full-stack feature audit across all layers (DB, API, auth, types, UI) |
/explain |
Plain-English code/concept explanation for any audience |
/compare |
Change visualization and impact assessment |
/dev-assess |
3-expert technical assessment with triangulation |
/adr |
Architecture Decision Record creation with conversational interview |
| Command | What It Does |
|---|---|
/feature-dev |
Complete feature development workshop (discovery, planning, build) |
/idea |
Feature brainstorming with approach exploration and feasibility |
/task |
Quick task creation with conversational details capture |
/prep |
Research and investigation document creation |
/reconcile |
Living document maintenance -- conversational drift detection |
| Command | What It Does |
|---|---|
/research |
Research orchestrator with queue, branch, and compound operations |
/distro |
Marketing/distribution pipeline (7 stages) |
/publish |
6-agent editorial pipeline for content publishing |
/write-article |
Write and publish articles with voice profile |
/write-research |
Generate research articles |
/post-linkedin |
LinkedIn content publishing with voice adaptation |
/announce-release |
Full release announcement pipeline |
| Command | What It Does |
|---|---|
/morning |
Daily standup brief across all active projects |
/status |
Quick project status check |
/save-state |
Checkpoint current work state |
/resume |
Resume from a saved checkpoint |
/review-codex |
Review and resolve multi-model builder work |
Specialized skills organized by domain:
| Category | Count | Examples |
|---|---|---|
| Frontend | 32 | ui-builder, nextjs-project-manager, layout-designer, expo-deployment |
| Backend | 19 | senior-backend, database-design, supabase-expert, api-design |
| Architecture & DevOps | 12 | senior-architect, senior-devops, cto-advisor, ci-cd-pipeline-builder |
| Code Quality | 10 | code-reviewer, testing-qa, test-generator, git-workflow-designer |
| AI & Automation | 15 | prompt-engineer, rag-pipeline-builder, agent-orchestrator, chain-builder |
| SEO | 6 | seo-analyst, seo-audit, programmatic-seo, schema-markup |
| CRO | 10 | form-cro, page-cro, signup-flow-cro, ab-test-designer |
| Content & Social | 20 | copywriting, blog-post-writer, social-content, newsletter-writer |
| Marketing Strategy | 9 | launch-strategy, growth-hacker, competitive-intelligence, gtm-strategist |
| Financial | 12 | cash-flow-forecaster, runway-calculator, revenue-modeler, budget-planner |
| Startup & Fundraising | 10 | pitch-deck-creator, fundraising-strategist, cap-table-manager |
| Operations | 8 | compliance-checker, contract-analyzer, vendor-manager |
| Product Management | 12 | mvp-planner, pmf-analyzer, user-research, roadmap-builder |
| Design | 10 | wireframe-creator, mockup-generator, brand-identity-builder |
| UX | 6 | onboarding-designer, chatbot-designer, faq-builder |
| Writing | 15 | technical-writer, white-paper-author, case-study-writer, grant-writer |
| Communication | 8 | email-composer, presentation-maker, slide-deck-designer |
| Browser Automation | 3 | browser-use, omni-vu, automation-architect |
Custom agents with tool access -- not static prompts, but executable specialists:
| Agent | Specialty |
|---|---|
| nextjs-senior-dev | Next.js 14+ App Router, performance, complex patterns |
| backend-architect | RESTful APIs, microservice boundaries, database schemas |
| operations-manager | Project coordination, quality control, process optimization |
| payment-security-specialist | Payment flow security, PCI compliance |
| stripe-integration-specialist | Stripe billing, subscriptions, webhooks |
| database-migration-specialist | Schema migrations, zero-downtime changes |
| mcp-server-architect | MCP server design and implementation |
| social-media-manager | Cross-platform content, engagement optimization |
| market-intelligence-analyst | Market research, trend analysis, competitor intelligence |
| text-editor-architect | Rich text editor design and implementation |
| steve-jobs-advisor | Product vision, design philosophy, strategic clarity |
| ...and 20+ more | Security, DevOps, testing, email, relationships |
Shell scripts for file-based project organization:
# Create workspace structure
./workspace/generators/new-project-workspace.sh
# Create task/bug/feature/ADR files
./workspace/generators/new-task.sh "Implement auth" --priority high
./workspace/generators/new-bug.sh "Login fails on Safari" --severity critical
./workspace/generators/new-feature.sh "Dark mode" --stage 5 --branch
./workspace/generators/new-adr.sh "Use Supabase over Firebase"
./workspace/generators/new-prep.sh "Auth options" --type tech-stack
# Complete a task (moves to .done/)
./workspace/generators/done-task.sh 2026-03-09-implement-auth.md./install.sh # Full install (commands + skills)
./install.sh --dry-run # Preview what will be installed
./install.sh --commands-only # Skip skills
./install.sh --uninstall # Remove all toolkit filesAll HTML visualizations follow the Factory-inspired design language:
| Token | Value | Usage |
|---|---|---|
| Background | #020202 |
Near-black |
| Text | #eeeeee |
Near-white |
| Accent | #ef6f2e |
Orange -- primary |
| Secondary | #f59e0b |
Amber -- secondary |
| Success | #4ecdc4 |
Teal -- success |
| Fonts | Geist + Geist Mono | With system fallbacks |
Rules: No shadows, no gradients, no glow effects. Typography and whitespace ARE the design.
squire.md and slash commands are prompt files. They don't execute code -- they instruct AI agents to use their built-in tools to scan, analyze, and generate. A well-specified prompt is a reusable tool.
Generator scripts are plain bash. They create markdown files with frontmatter metadata. No dependencies beyond bash 3.2+.
Templates are document starters. Copy them or let /reconcile init generate them conversationally.
This toolkit was built by Eddie Belaval at id8Labs while shipping 12+ AI-augmented products with Claude Code over 5 months. Every rule, pattern, and template came from real friction -- sessions where something went wrong, and the fix was documented so it wouldn't happen again.
| Metric | Count |
|---|---|
| Sessions | 2,990 |
| Commits | 3,307 |
| Products shipped | 12+ |
| Skills built | 327 |
| Slash commands | 63 |
| Custom agents | 34 |
| Duration | 5 months (Oct 2025 - Mar 2026) |
Oct 2025: ████████████████░░░░░░░░░░░░░░ 364 commits (ramp-up)
Nov 2025: ███████████░░░░░░░░░░░░░░░░░░░ 269 commits (learning)
Dec 2025: █████████░░░░░░░░░░░░░░░░░░░░░ 215 commits (holidays)
Jan 2026: ██████████████████████████████░ 916 commits (ignition)
Feb 2026: ████████████████████████████████ 1,215 commits (peak)
Mar 2026: █████████████████████░░░░░░░░░░ 487 commits (12 days in)
The behavioral rules are backed by data: the top friction patterns across those sessions became the 9 rules. The pipeline came from shipping 12 products through the same stage-gate system. The thinking frameworks came from doing the same types of reviews repeatedly and wanting consistency.
If this helps you build better with AI, that's the goal.
X: @eddiebe | GitHub: eddiebelaval | Site: id8labs.app
MIT License. See LICENSE for details.
