etyb.ai  ·  Changelog  ·  Stacks  ·  Install Guide  ·  Architecture

etyb-skills

Install a virtual engineering company. One slash command — /etyb — gives your AI agent a CTO, 20 specialists, 9 always-on engineering disciplines, and platform-specific Stack Packs.

v4 collapses what used to be 30 separate skills into one coordinated skill with currency-stamped vendor knowledge living inside the repo at stacks/<vendor>/. The user only ever invokes /etyb. ETYB silently routes the work to the right internal specialist, applies the right protocols, reads the relevant Stack page directly when the work involves a vendor, and signs every response. New: the public changelog is linked under every response — one channel, no slash-command pollution.

Quick Start

# Any agent — Claude Code, Codex, Cursor, Kiro, Trae, Antigravity
npx skills add e-t-y-b/etyb-skills

Or, on Claude Code, install as a native plugin instead:

/plugin marketplace add e-t-y-b/etyb-skills
/plugin install etyb@etyb-skills

Once installed, every request goes through /etyb. Specialist expertise is loaded internally as you need it.

Platform Support

Who This Is For

• Teams shipping real software — your AI agent works like a disciplined team member, not a solo cowboy
• Engineers tired of AI "yes-and" behavior — agents that skip tests, rubber-stamp reviews, or chase symptoms
• Regulated or high-stakes codebases (fintech, healthcare, e-commerce) — traceable decisions, gated releases, evidence-backed claims
• Solo developers — /etyb is one slash command and works the same regardless of project size

What You Get

An AI coding agent that works like a 100-person engineering org — through a single trigger:

• Refuses to ship untested code — TDD-first discipline, backed by advisory hooks on the Claude Code plugin path (warn, never block)
• Stops you from building the wrong thing — structured brainstorming before architecture
• Pushes back on bad review feedback — evaluates findings on merit, no performative agreement
• Coordinates parallel work — subagent dispatch with two-stage review and worktree isolation; heavy work runs in sub-agent contexts, model-tiered (cheap models for cheap work, never above your own selected model), so your session stays fast and your context stays clean
• Covers the full SDLC — from research through production operations
• Knows your domain — fintech ledgers, HIPAA compliance, e-commerce patterns, real-time systems
• Speaks your platform — Stack Packs load across all roles when work involves a specific stack. 13 Stacks are live under stacks/: AWS, GCP, Azure, Salesforce, Anthropic Claude, OpenAI, Cloudflare, Vercel, Supabase, Firebase, Expo, Stripe, Observability
• Identifies itself — every Tier 1-4 response ends with ETYB · <role-engaged> and a What's new — etyb.ai/changelog line. One brand, transparent expertise.

Architecture

USER REQUEST
     │
   /etyb (the only trigger surface — single skill, single brand)
     │
ETYB CTO core ── reads internal references on demand:
     │
     ├── references/specialists/  (14 core engineering team READMEs)
     ├── references/protocols/    (9 always-on engineering disciplines)
     ├── references/verticals/    (6 business-domain architects)
     │
     │  ALWAYS-ON PROTOCOL LAYER (loaded into every response)
     │  ├── TDD — no code without failing test
     │  ├── Verification — evidence before claims
     │  ├── Review — no performative agreement
     │  ├── Plan execution — one task at a time
     │  ├── Brainstorm-first — explore before solving
     │  ├── Branch safety — never merge without green tests
     │  ├── Subagent coordination — parallel dispatch + review
     │  ├── Self-improvement — failing eval before skill changes
     │  └── Debugging — root-cause-first after repeated failures
     │
     │  STACK PACK OVERLAYS (load when platform signals match)
     │  └── 13 vendors — AWS, GCP, Azure, Salesforce, Anthropic, OpenAI,
     │      Cloudflare, Vercel, Supabase, Firebase, Expo, Stripe, o11y
     │
RESPONSE — signed: ETYB · <role-engaged> + changelog link

Three orthogonal axes: specialists (the roles), protocols (the disciplines), Stack Packs (the platforms). All live as internal references under skills/etyb/references/. Adding a new stack adds a folder, not a roster slot — and never a new slash command.

What's Inside ETYB

Specialists (14, under references/specialists/)

Verticals (6, under references/verticals/)

Protocols (9, under references/protocols/)

Stack Packs

Platform-specific knowledge overlays with knowledge-currency timestamps, authoritative-source URLs, per-product drift-risk ratings, and vendor-skill delegation — when a vendor MCP/skill is installed in your environment, ETYB defers to it rather than answering from baked knowledge. The team doesn't grow — it learns the platform.

13 Stacks live in this repo under stacks/ with a unified knowledge-currency framework. ETYB reads per-product and per-role pages directly when the install is on disk. Third-party agents without the install can fetch the same content as raw markdown from https://raw.githubusercontent.com/e-t-y-b/etyb-skills/main/stacks/<vendor>/<page>.md. See STACKS.md for the full registry, drift-risk model, and authoring conventions.

Roadmap: Snowflake, Databricks, dbt, Shopify, SAP, ServiceNow, Twilio, Auth0/Okta/Clerk/WorkOS as separate Stacks.

Install

Skills CLI (any agent)

npx skills add e-t-y-b/etyb-skills

The skills CLI installs the full /etyb skill — 14 specialists + 9 protocols + 6 verticals — into every detected agent's skills directory (Claude Code .claude/skills/, Codex/Antigravity/Trae .agents/skills/, Kiro .kiro/skills/, Cursor .cursor/skills/). Project installs record a skills-lock.json — commit it so your team restores the same skill set with npx skills experimental_install.

Claude Code (plugin)

/plugin marketplace add e-t-y-b/etyb-skills
/plugin install etyb@etyb-skills

Native plugin install from this repo's .claude-plugin/ manifest. Use the skills CLI or the plugin, not both. See docs/installation.md for per-harness details, verification, and troubleshooting.

Enforcement note: on the Claude Code plugin path, five advisory hooks (hooks/hooks.json) and five specialist agent definitions (agents/) ship with the install — the hooks surface warnings (no test file before an edit, no review evidence before a commit, merging without green tests) but never block. On skills-CLI installs for other harnesses, ETYB's disciplines are model-trusted — enforced by instruction only.

Updating

npx skills update

Every ETYB response ends with What's new — etyb.ai/changelog so you always have a one-click path to release notes. On the plugin path, update the marketplace and reinstall from the /plugin menu.

How Skills Load (Token Efficiency)

Skills use progressive disclosure — a markdown-based RAG pattern:

Layer 0  Runtime guardrails (0 tokens — scripts outside the LLM)
Layer 1  ETYB SKILL.md always loaded (~800 tokens — the culture)
Layer 2  Relevant reference loads on demand (~2,500 tokens — the specialist)
Layer 3  Deep reference loads on demand (~4,000 tokens — single helper file)

Per-activation: ~6,000-10,000 tokens (not the whole library)
Tier 0-1 requests: ~3,500 tokens (ETYB handles directly)

The v4 collapse cuts metadata cost — instead of 30 always-loaded descriptions competing at trigger time, there's one. Internal references are loaded only when ETYB consults them.

Evidence: With vs Without

Average token overhead: +53%. What you get: an engineer who says "no" when it matters.

Links

• Website: etyb.ai
• Changelog: etyb.ai/changelog
• Repo: github.com/e-t-y-b/etyb-skills
• Install guide: docs/installation.md
• Architecture: docs/architecture.md
• Issues & contributions: github.com/e-t-y-b/etyb-skills/issues

License

MIT — see LICENSE.