What's New
Skills 17–19 + integration arc (2026-06)
- Skill 17 — Non-Engineering Verticals — C-suite personas, finance domain, compliance OS, PM domain. One OS, every function.
- Skill 18 — Document Processing — DOCX/XLSX read-write, PDF generation + parsing, PPTX generation. Full pipeline, no manual exports.
- Skill 19 — MCP Authoring — forge HTTP MCP servers on Workers, stdio MCP templates, OpenAPI→MCP forger. Publish to the registry in one command.
- 20+ slash commands —
/saas, /forge-from-openapi, /forge-webhook-handler, /generate-prp, /execute-prp, /run-evals, /audit-doctrine, /audit-mcp-fleet, /dashboard-cockpit, /drift-check, and more.
- Production forge script —
bin/forge-skill-from-openapi.mjs generates a full SKILL.md + command scaffold from any OpenAPI spec.
- LLM eval harness —
/run-evals + structured rubrics + schema-validated results + regression tracking. AI-heavy behavior now has a repeatable quality gate.
- 15 new doctrinal principles —
state-is-the-enemy, fail-fast-build-fail-soft-prod, ttfr-north-star, sync-ui-async-backing, cost-per-request-accountability, inverted-abstraction-pyramid, one-way-two-way-doors, vendor-risk-tiering, data-residency-by-default, right-to-deletion, refund-automation, documentation-as-code, production-observability-default-on, hardware-aware-programming, webhook-receiver-architecture.
- DO agent scaffolding — Durable Objects as stateful AI agents via skill 19, wired to the MCP registry.
- RAG pipeline — vectorize + D1 hybrid search documented in skill 05 architecture submodules.
- Multi-tenant subdomain provisioning — Cloudflare custom hostnames + per-tenant D1 namespacing, baked into skill 05.
- 5 new PreToolUse hooks — permission-discipline enforcement before destructive tool calls.
- MCP server registry —
/audit-mcp-fleet catalogs, health-checks, and version-pins every MCP in your harness.
When to Use
| Building... |
Skills Load |
Install |
| SaaS product |
01-OS → 02-Brief → 05-Arch → 06-Build → 07-QA → 08-Deploy + all agents |
claude plugin install heymegabyte/claude-skills |
| API / backend |
01-OS → 05-Arch → 06-Build → 08-Deploy → 13-Observability |
npm i @heymegabyte/claude-skills |
| Marketing site |
01-OS → 09-Brand → 10-Design → 11-Motion → 12-Media |
npm i @heymegabyte/claude-skills |
| Features on existing project |
01-OS → 06-Build → 07-QA (skill router auto-selects from context) |
claude plugin install heymegabyte/claude-skills |
| Just rules for your AI tool |
No skills — use a platform variant file directly |
See 32 variants below |
How It Works
┌─────────────────────────────────────────────────────────────────┐
│ YOUR PROMPT │
│ "Build a SaaS for dog walkers" │
└───────────────────────────┬─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ SKILL ROUTER │
│ Matches prompt → loads smallest useful subset of skills │
│ 01-OS always loaded │ then 02-Brief → 05-Arch → 06-Build │
└───────────────────────────┬─────────────────────────────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────┐
│ ARCHITECT│ │ PARALLEL │ │ PARALLEL │
│ (Opus) │ │ BUILD │ │ VERIFY │
│ │ │ 3-5 agents│ │ 3 agents │
│ repo-map │ │ frontend │ │ deploy │
│ task graph│ │ backend │ │ seo │
│ seams │ │ content │ │ visual │
└──────────┘ │ media │ │ a11y │
│ tests │ └──────────┘
└──────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ HARD GATES │
│ Playwright 6bp ✓ │ Vision ≥8/10 ✓ │ Lighthouse A11y ≥95 ✓ │
│ axe-core 0 ✓ │ SEO GREEN ✓ │ 0 errors ✓ │ Flesch ≥60 ✓ │
└───────────────────────────┬─────────────────────────────────────┘
│
▼
┌──────────────┐
│ DEPLOYED │
│ CF Workers │
│ + purged │
└──────────────┘
Install
# GitHub Skills (recommended)
gh skill install heymegabyte/claude-skills
# Claude Code plugin
claude plugin install heymegabyte/claude-skills
# npm (includes all 32+ platform variants)
npm i @heymegabyte/claude-skills
# JSR
npx jsr add @heymegabyte/claude-skills
# OpenAI Codex
git clone https://github.com/heymegabyte/claude-skills ~/.codex/skills
# Manual
git clone https://github.com/heymegabyte/claude-skills ~/.agentskills
Cross-Platform Support (32 variants)
Auto-generated on every push to master. Each format uses native frontmatter for its tool.
| Format |
Path |
Notes |
| Cursor (modern) |
.cursor/rules/emdash-skills.mdc |
MDC frontmatter: alwaysApply: true |
| Windsurf (modern) |
.windsurf/rules/emdash-skills.md |
trigger: always_on |
| Augment (modern) |
.augment/rules/emdash-skills.md |
type: always_apply |
| Copilot (modern) |
.github/instructions/emdash-skills.instructions.md |
applyTo: "**" |
| OpenHands |
.openhands/microagents/repo.md |
Plain markdown |
| Cursor (legacy) |
.cursorrules |
Single-file format |
| Windsurf (legacy) |
.windsurfrules |
Single-file format |
| Cline |
.clinerules |
Single-file format |
| Copilot (legacy) |
.github/copilot-instructions.md |
Plain markdown |
| Augment (legacy) |
.augment-guidelines |
Single-file format |
| Aider |
.aider-conventions.md |
Plain markdown |
| Zed |
.rules |
Plain markdown |
| Codex |
CODEX.md + .agents/skills/ |
SKILL.md per category |
| Gemini CLI |
GEMINI.md |
Plain markdown |
| Amp |
AMP.md |
Plain markdown |
| Replit |
replit.md |
Plain markdown |
| Devin |
.devin/skills/emdash/SKILL.md |
SKILL.md with frontmatter |
| Goose |
.goose/recipes/emdash-skills.yaml |
YAML recipe format |
| AGENTS.md |
AGENTS.md |
Devin, Jules, Copilot, OpenHands |
| Amazon Q |
.amazonq/rules/ |
Directory format |
| JetBrains Junie |
.junie/guidelines.md |
Plain markdown |
| Trae |
.trae/rules/project_rules.md |
ByteDance IDE |
| Tabnine |
.tabnine/guidelines/guidelines.md |
Directory format |
| Kilo Code |
.kilo/rules/ |
Roo Code successor |
| Roo Code |
.roo/rules/ |
Active until May 2026 |
| Continue.dev |
.continue/rules/ |
Directory format |
| JetBrains AI |
.aiassistant/rules/emdash-skills.md |
Separate from Junie |
| Kiro (AWS) |
.kiro/steering/emdash-skills.md |
Workspace steering |
| Void |
.void/rules/emdash-skills.md |
Open-source VS Code fork |
| Qodo |
QODO.MD |
PR-Agent auto-detects |
| Bolt.new |
.bolt/promptfile |
Plain text instructions |
| Cursor BugBot |
.cursor/BUGBOT.md |
Per-directory code review |
Skill Categories
┌──────────────────────────────────────────────┐
│ EMDASH SKILL MAP │
│ │
│ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ │
│ │ 01 │→ │ 02 │→ │ 03 │→ │ 04 │ │
│ │ OS │ │Brief│ │Plan │ │Pref │ │
│ │ 6 │ │ 0 │ │ 1 │ │ 3 │ │
│ └─────┘ └─────┘ └─────┘ └─────┘ │
│ │
│ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ │
│ │ 05 │→ │ 06 │→ │ 07 │→ │ 08 │ │
│ │Arch │ │Build│ │ QA │ │Ship │ │
│ │ 12 │ │ 26 │ │ 25 │ │ 10 │ │
│ └─────┘ └─────┘ └─────┘ └─────┘ │
│ │
│ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ │
│ │ 09 │→ │ 10 │→ │ 11 │→ │ 12 │ │
│ │Brand│ │ UX │ │Motn │ │Media│ │
│ │ 7 │ │ 2 │ │ 1 │ │ 10 │ │
│ └─────┘ └─────┘ └─────┘ └─────┘ │
│ │
│ ┌─────┐ ┌─────┐ ┌─────┐ │
│ │ 13 │→ │ 14 │→ │ 15 │ │
│ │Grow │ │Ideas│ │ Gen │ │
│ │ 7 │ │ 0 │ │ 9 │ │
│ └─────┘ └─────┘ └─────┘ │
│ │
│ Numbers = reference docs per category │
└──────────────────────────────────────────────┘
| # |
Category |
Docs |
What It Handles |
| 01 |
Operating System |
6 |
Core policy, autonomy, parallelization, AI-native coding, output compression |
| 02 |
Goal & Brief |
— |
Product thesis from a domain name, business model inference |
| 03 |
Planning & Research |
1 |
Competitive analysis, task decomposition, parallel workstreams |
| 04 |
Preference & Memory |
3 |
Voice of Customer data, user preferences, behavioral psychology |
| 05 |
Architecture & Stack |
12 |
CF Workers, Hono, Drizzle v1, Coolify, MCP, auth, API design, multi-tenancy |
| 06 |
Build & Slice Loop |
26 |
Forms, search, blog, i18n, PWA, webhooks, admin, data tables, chat |
| 07 |
Quality & Verification |
25 |
E2E, a11y, security hardening, perf, visual QA, adversarial, AI testing |
| 08 |
Deploy & Runtime |
10 |
CI/CD, launch-day sequence, uptime, backup, changelog, GitHub CI fix |
| 09 |
Brand & Content |
7 |
SEO engine, copy system, email templates, social automation |
| 10 |
Design System |
2 |
Anti-slop design, dark-first, bold typography, CSS architecture |
| 11 |
Motion & Interaction |
1 |
Scroll-driven animations, View Transitions, reduced-motion |
| 12 |
Media Orchestration |
10 |
Image/video generation, AI prompts, compression, OG previews, diagramming |
| 13 |
Growth & Observability |
7 |
Stripe billing, analytics, Sentry alerts, email, experiments, CRO |
| 14 |
Idea Engine |
— |
Autonomous research, evidence-backed improvement proposals |
| 15 |
Site Generation |
9 |
Full-corpus rebuild, slot-first GPT Image 1.5 media, NotebookLM podcast/infographic/video, citations |
| 16 |
Cinematic Website |
— |
100-rule prime directive: cinematic UX, PWA, JSON-LD, CWV, a11y, integrations, TDD, deploy |
| 17 |
Non-Engineering Verticals |
4 |
C-suite personas, finance domain, compliance OS, PM domain — full-org coverage |
| 18 |
Document Processing |
4 |
DOCX/XLSX read-write, PDF parse + generate, PPTX generation — zero manual exports |
| 19 |
MCP Authoring |
3 |
HTTP MCP on Workers, stdio server templates, OpenAPI→MCP forger, registry publish |
Agents
AGENT ROUTING BY MODEL TIER
┌─────────────────────────────────────┐
│ OPUS (heavy) │
│ Architecture │ Security │ Vision │
│ Completeness │ Meta-orchestration │
├─────────────────────────────────────┤
│ SONNET (standard) │
│ Build │ Test │ Deploy │ Debug │
│ Migrate │ Profile │ Simplify │
│ Dependencies │ Incidents │ CU-op │
├─────────────────────────────────────┤
│ HAIKU (fast) │
│ Content │ SEO │ A11y │ Changelog │
│ Cost estimation │
└─────────────────────────────────────┘
| Agent |
Model |
Effort |
Purpose |
| architect |
Opus |
max |
Repo-map generation, task graphs, architectural seams |
| completeness-checker |
Opus |
max |
Zero Recommendations Gate — nothing ships incomplete |
| meta-orchestrator |
Opus |
max |
Cross-agent coordination and task graph execution |
| security-reviewer |
Opus |
max |
OWASP Top 10:2025, secrets exposure, CSP audit |
| visual-qa |
Opus |
max |
Screenshot all 6 breakpoints, AI vision layout detection |
| code-simplifier |
Sonnet |
high |
Reduce complexity while preserving all functionality |
| computer-use-operator |
Sonnet |
high |
Native macOS app automation via Computer Use MCP |
| dependency-auditor |
Sonnet |
high |
Package security, license compliance, update strategy |
| deploy-verifier |
Sonnet |
high |
Post-deploy smoke tests at 6 breakpoints |
| incident-responder |
Sonnet |
high |
Sentry-triggered triage, root cause, auto-fix PRs |
| migration-agent |
Sonnet |
high |
Framework/DB/API migration with rollback safety |
| performance-profiler |
Sonnet |
high |
Core Web Vitals analysis, bundle audit, runtime profiling |
| test-writer |
Sonnet |
high |
TDD-first Playwright E2E + Vitest units, stable selectors |
| accessibility-auditor |
Haiku |
low |
axe-core WCAG 2.2 AA audits and remediation guidance |
| changelog-generator |
Haiku |
low |
Conventional commit parsing, user-outcome release notes |
| content-writer |
Haiku |
low |
Marketing copy, microcopy, alt text, SEO content |
| cost-estimator |
Haiku |
low |
Cloudflare Workers cost forecasting and free-tier warnings |
| seo-auditor |
Haiku |
low |
Title, meta, H1, JSON-LD, OG tags, sitemap validation |
| notebooklm-orchestrator |
Sonnet |
high |
Podcast (ElevenLabs) + infographic + explainer video pipeline per site |
Hard Gates
Every deploy must clear all gates. No exceptions. No overrides.
| Gate |
Threshold |
Tool |
| E2E Tests |
0 failures @ 6 breakpoints |
Playwright v1.59+ |
| Visual QA |
≥ 8/10 |
AI vision (Claude Sonnet 4.6 default; current OpenAI multimodal fallback) |
| Accessibility |
≥ 95 |
Lighthouse |
| A11y Violations |
0 |
axe-core |
| SEO Score |
GREEN |
Yoast-equivalent |
| Console Errors |
0 |
Browser DevTools |
| Placeholders |
0 |
Content sweep |
| Readability |
Flesch ≥ 60 |
Copy audit |
Stack
REQUEST FLOW
Browser ──→ CF Workers ──→ Hono RPC ──→ Drizzle v1 ──→ D1/Neon
│ │ │ │
│ KV/Upstash Zod valid. Migrations
│ │ │
├── Clerk (auth) Turnstile
├── Stripe (pay) Resend (email)
├── PostHog (analytics) Sentry (errors)
└── GA4/GTM (tracking) Inngest (jobs)
| Layer |
Technology |
| Hosting |
Cloudflare Workers |
| Backend |
Hono RPC + @hono/zod-validator |
| Frontend |
Angular 21 + Ionic 8 + PrimeNG 21 (or vanilla) |
| Database |
D1 (edge) / Neon (Postgres) |
| ORM |
Drizzle v1 + Zod |
| Cache |
KV / Upstash Redis |
| Auth |
Clerk |
| Payments |
Stripe |
| Email |
Resend + Listmonk |
| Jobs |
Inngest |
| Testing |
Playwright v1.59+ + Vitest |
| Lint |
ESLint + Prettier |
| Runtime |
Bun |
| Monitoring |
PostHog + Sentry + GA4/GTM |
Templates
| Template |
Purpose |
acceptance-criteria.md |
Structured AC with testable conditions |
adr-template.md |
Architecture Decision Records |
brief.md |
Product brief from domain name |
launch-checklist.md |
Pre-launch verification checklist |
product-intent.yaml |
Machine-readable product definition |
qa-report.json |
Structured QA output format |
repo-map.md |
Codebase architecture map |
saas-feature-manifest.md |
Complete SaaS feature matrix |
session-learning.md |
Post-session knowledge extraction |
starter-scaffold.md |
New project scaffolding guide |
semgrep-rules/ |
Custom Semgrep rules for codebase enforcement |
task-graph.json |
Parallelizable task decomposition |
Tools
| Script |
Purpose |
bin/check-required-keys.sh |
API Key Gate helper — verifies per-mode required keys present, JSON output + audit log |
bin/emdash-quick-lint |
Staged-file linting (ESLint+Prettier+ShellCheck) |
bin/emdash-secret-scan |
Pre-commit secret detection via detect-secrets |
bin/emdash-cost-check |
Quick Cloudflare cost estimate from wrangler.toml |
scripts/discover-secrets.sh |
Inventory all available API keys across sources |
scripts/gpt4o-vision-analyze.sh |
OpenAI multimodal screenshot analysis for visual QA (script filename predates the 2026-02 GPT-4o retirement; uses current GPT Image 2 vision per platform.openai.com/docs/deprecations) |
scripts/validate-skills.sh |
Cross-reference validation for skill files |
scripts/visual-tdd-loop.sh |
Automated screenshot→fix→verify loop |
Task Routing
The router loads the smallest useful subset per task — never the full 163 docs. See _router.md for the complete routing table. This plugin ships 163 reference docs across 19 skill categories.
| When you say... |
Skills loaded |
| "Build a new project" |
02 → 03 → 05 → 06 → 09 |
| "Add a feature" |
05 → 06 → 07 |
| "Fix CI" |
07 → 08 (especially gh-fix-ci) |
| "Deploy this" |
08 (+ 09 if content changed) |
| "Polish the frontend" |
09 → 10 → 11 → 12 |
| "Set up billing" |
05/auth → 06/webhooks → 13/stripe |
| "Add analytics" |
13 (+ 09/social if publishing) |
| "Brainstorm ideas" |
03 → 14 |
Ideal Prompts
Copy-paste these as-is. Each one routes through the skill graph and produces a deployed, gate-cleared product. The shorter the prompt, the more the skill engine infers — domain name alone is enough.
Build a SaaS from a domain
Build acmebilling.dev — recurring subscription billing for indie SaaS founders. Stripe-native, Clerk auth, Inngest for retry workflows.
- Skills — 01 → 02 → 03 → 05 → 06 → 07 → 08 → 13
- Agents — architect (Opus) → 5 parallel build agents → deploy-verifier + seo-auditor + visual-qa
- Gate fires — API Key Gate checks
STRIPE_*, CLERK_*, INNGEST_*, CLOUDFLARE_API_KEY (global) before any scaffold
- Output — live Worker on
acmebilling.dev with checkout, dashboard, webhook handler, Sentry + PostHog + GA4 wired
Rebuild an existing site
Rebuild brianzalewski.com — pull current site from Wayback if dead, keep every page, modernize design, ship to CF Workers.
- Skills — 01 → 02 → 09 (brand extraction) → 15 (full corpus + media+video extraction + grammar audit) → 07 → 08
- Agents — architect → content-writer + visual-qa + seo-auditor
- Gate — every original URL = 200 or 301; logo retention; theme match; favicon set via real-favicongenerator
Spin up a portfolio from one line
Make me a portfolio site at brian.dev — I'm a principal engineer, AI builder, Megabyte Labs founder.
- Skills — 01 → 02 (founder inference) → 09 → 10 → 11 → 12 → 14 (auto-suggests project tiles + impact stats) → 07 → 08
- Optional pairing — if a flagship SaaS exists, portfolio links to it as primary work
- Output — dark-first single-page with hero + work + writing + contact
Local-business site (NAP + reviews + map)
Build paterson-dental.com — family dental practice in Paterson NJ, online booking, insurance verification, Spanish/English.
- Skills — 01 → 02 → 09 → 10 → 13 (local conversion patterns: phone_click, direction_click, booking_click)
- Gate —
GOOGLE_MAPS_API_KEY + GOOGLE_PLACES_API_KEY required
- Output — full-width Maps embed, NAP schema.org
LocalBusiness, OpenTable/Booksy embed, Twilio SMS booking
Non-profit with donation flow
Build sjsk.org — community clothing distribution for SJSK in Newark, donor portal, impact counter, tax receipt PDFs.
- Skills — 01 → 02 → 09 → 13 (Stripe-first GiveDirectly UX)
- Gate —
STRIPE_* keys
- Output — Donate CTA in nav, recurring + one-time, Resend tax receipts, impact counters with IO+rAF roll-in
Brainstorm before you build
What's the highest-ROI thing I can ship this week? Read PORTFOLIO.md, scan current projects, propose 3 ideas with confidence scores.
- Skills — 14 (idea engine) → bounded web research → self-critique filter
- Output — 3 evidence-backed proposals with
apa_citation per claim, viral coefficient + AI search visibility scoring, auto-implements confidence ≥ 0.85 aligned ideas
Add a feature to an existing project
Add a magic-link auth flow to acme.dev — passwordless email via Resend, Clerk session, redirect to /dashboard.
- Skills — 05 (architecture decision) → 06 (build slice) → 07 (Playwright E2E homepage-first)
- Agents — test-writer (failing test FIRST) → implementation → deploy-verifier
Debug a wedged pipeline
projectsites.dev workflow stuck on site_id 47 for 3 hours. Diagnose, fix, retrigger.
- Skills — 08 (deploy/runtime) → 07
- Rule loaded —
failed-pipeline-protocol.md (5 canonical failure modes)
- Sequence — detect via D1 query → diagnose root cause (CHECK constraint? timeout? OOM?) → fix → verify in isolation → mint session → retrigger via direct worker URL → background monitor
Ship a content-driven blog at scale
Generate 50 programmatic SEO pages for acme.dev — integration|comparison|use-case|template|location templates, GEO-optimized, citations.
- Skills — 09 (pSEO 5 types + GEO + sourced facts via APA citations) → 06 → 07
- Output — 50 unique routes, no templated copy, every quantitative claim cites APA 7th source, Schema.org
Article with citation:CreativeWork[]
One-line magic
ghost.megabyte.space
- Skills — 01 (one-line prompt mode inference) → 02 → 05 → 09 → 10 → 11 → 12 → 07 → 08
- Domain alone routes to a complete product
- Gate fires
- Brand extracted from existing infrastructure
- Curated dark/neon aesthetic from
~/Snapchat/best/ 622-shot reference
- Logo from Ideogram
- Deployed to CF Workers
Prompt patterns that work
| Phrase |
Effect |
boil the lake |
Force complete-not-shortcut mode for the next decision |
parallel everything |
Decompose first, spawn 3-5 agents per phase |
skip api key gate |
Bypass key check (rare — deploys will likely 500) |
recommendations loop until zero |
Keep finding+fixing until no rec remains |
chain MCPs |
Use meta-orchestrator across 19+ services |
homepage-first |
Reset E2E flow to start at / and click through |
boil X, flag Y |
Do X completely; surface Y as issue |
make it shorter |
Reduce 40-60% (Brian's #1 most-used phrase, 670+ logged uses) |
the whole thing |
Never truncate — full file output |
Now ... |
Chain to previous task, don't reset context |
Philosophy
- Distribution > Technology — The best tool nobody knows about is the worst tool. Auto-create repos for new skills. Integrate into every ecosystem. Broadcast widely.
- Boil the Lake — When completeness costs minutes more than a shortcut, do complete. Boil lakes, flag oceans.
- TDD Always — Failing test first → implement → pass. Real user flows. Homepage first. Click through UI. Never
page.goto() for internal navigation.
- One Person + AI = Twenty — The barrier is gone. What remains is taste, judgment, and willingness to do the complete thing.
Can You Make This Better?
Seriously — open an issue or submit a PR. Some things we're thinking about:
- More skill categories? Is 15 the right number or are we missing something?
- Better agent routing? Should model assignments shift as Claude evolves?
- Templates you wish existed? What boilerplate do you write over and over?
- Skills for other stacks? This is CF Workers + Angular today. What else?
- Prompting patterns that consistently produce better results?
If you've built something similar, stolen ideas from here, or just have opinions — we want to hear it. The whole point is that this gets better every day.
License
Copyright (c) 2024-2026 Brian Zalewski / Megabyte Labs. The Rutgers License.
TL;DR — It's free. Use it. But if it helped you, be cool about it and send what feels right. We made this and we'd like to eat.
No comments yet
Be the first to share your take.