References — full external index for Etz Chaim AI¶

Every link Claude Code / Codex / Cursor / Gemini may need to fetch in order to operate effectively on this codebase. Organized by topic.

Tip for agents: prefer fetching these references when you encounter a workflow / tool / pattern question. If the reference is fresh enough, follow it; if it's superseded, the most recent version usually links from the cited page.

I. Boris Cherny — creator of Claude Code¶

Primary sources¶

Boris's pinned thread (Jan 2, 2026) — the 8M-view "watershed moment" workflow disclosure
Worktree feature announcement (Feb 20, 2026) — --worktree flag, isolation: worktree subagents
/batch skill announcement (Feb 27, 2026) — parallel migration orchestration

Curation & analysis¶

howborisusesclaudecode.com — full curated tip compendium
VentureBeat manifesto
Get Push To Prod breakdown
Karo Zieminski substack breakdown
MindwiredAI 7 secrets

Key principles (from these sources)¶

~100 lines / ~2500 tokens for CLAUDE.md
150-200 instruction budget total
5 parallel terminal Claudes + 5-10 browser Claudes
Plan Mode (Shift+Tab×2) before all PRs
Verification loop = 2-3× quality boost
Every mistake becomes a rule
Slash commands for every workflow you do >1×/day
Worktrees for parallel agents
PostToolUse formatting hook (the last 10%)
agent-stop hook for deterministic verification

II. Anthropic engineering — foundational papers¶

Architecture & context¶

Effective context engineering for AI agents (Sep 2025) — attention budget, compaction, structured note-taking, multi-agent
Effective harnesses for long-running agents — initializer + coding agent + persistent artifacts (directly applicable to our daemon)
How we built our multi-agent research system — multi-agent patterns, evaluator-optimizer
Building Effective Agents (Erik Schluntz)
Building Effective AI Agents (Brendan Falk's analysis) — sequential, parallel, evaluator-optimizer patterns

Tools & evals¶

Writing tools for agents (Ken Aizawa) — the gold standard for MCP tool descriptions
Demystifying evals for AI agents — 20-50 tasks to start, Swiss Cheese model
Anthropic Cookbook — building_evals.ipynb — concrete eval recipes

Skills¶

Equipping agents with Agent Skills — the foundational essay
anthropics/skills GitHub — official reference skills
Agent Skills standard (cross-tool)

Auto Mode & sandboxing¶

Claude Code Auto Mode — Sonnet 4.6 classifier + prompt-injection probe
Beyond Permission Prompts: Making Claude Code More Secure — sandbox-based security, 84% prompt reduction

III. Anthropic thesis on agents — Barry Zhang & Mahesh Murag¶

The "Don't build agents, build skills instead" position.

Talk on YouTube (16 min) — the canonical talk
Class Central summary
Fresh HQ analysis
Bagrounds key takeaways
Lilys notes (detailed)
Barry Zhang bio — author of Skills system, key principles distilled
Mahesh Murag's Twitter — for follow-ups

IV. Erik Schluntz — head of programming agents¶

Author of Building Effective Agents (Anthropic 2024-2025).

Key principles¶

15-20 min upfront with Claude exploring the codebase before writing any code
Then one consolidated prompt
Treat yourself as the PM of Claude
22 000-line PR merged in production RL codebase via this pattern
Plan mode as the speccing tool

V. Code with Claude SF — May 6, 2026¶

The Anthropic developer conference.

Official event page (SF)
Conference index page — also lists London May 19 and Tokyo June 10
Blake Crosley recap (deep dive)
Chris Ebert notes (in-person attendee)
Context Studios field guide
Generation Digital preview
Pasquale Pillitteri preview
GadgetBond preview
Releasebot updates — Anthropic May 2026

VI. Claude Code documentation¶

VII. Anthropic API documentation¶

VIII. Standards & cross-tool¶

agentskills.io — Agent Skills standard
agents.md — AGENTS.md standard (Linux Foundation AAIF)
modelcontextprotocol.io — MCP spec
Linux Foundation AAIF — governance
containers.dev — Dev Containers spec
opentelemetry.io — OpenTelemetry
json-schema.org — JSON Schema

IX. Distribution & marketplaces¶

skills.sh — Vercel package manager for skills
skillsmp.com — community skills marketplace
agensi.io — security-scanned skills
mcp.so — MCP server registry
Glama — MCP discovery
Smithery — MCP server marketplace
buildwithclaude.com — Claude Code plugin marketplace
awesome-agent-skills — community-curated list
awesome-claude-code — Claude Code ecosystem
awesome-codex-cli — Codex CLI ecosystem

X. Cross-provider tooling¶

LiteLLM (BerriAI) — 100+ providers
LiteLLM docs
Pydantic AI — typed multi-provider agent framework
Pydantic AI Gateway — zero-translation routing
OpenCode — 150K stars MIT, 75+ providers via Models.dev
Models.dev — provider registry
Bifrost (Maxim AI) — Go-based gateway, 11μs overhead

XI. DevContainer references¶

XII. Comparative tooling analysis¶

For understanding the Claude Code vs Codex CLI vs Cursor vs OpenCode landscape:

Codex vs Claude Code (Blake Crosley)
Codex vs Claude Code (Builder.io)
Codex vs Claude Code (Developers Digest)
Codex vs Claude Code (Codersera)
Claude Code vs Codex CLI (NxCode)
OpenCode vs Codex vs Claude (Nimbalyst)
MindStudio analysis
VILA-Lab — Dive into Claude Code (paper) — comprehensive academic analysis

XIII. Patterns & community¶

Self-healing agent loops (MindStudio) — directly applicable to our Karpathy daemon
Adversarial PR review (MindStudio) — Claude proposes, Codex reviews
Git worktrees for parallel AI (Botmonster)
Mastering Git worktrees with Claude Code (Tuna)
Parallel vibe coding (Dan Does Code)
Compound Engineering plugin
Ralph Wiggum loop pattern
BigBrain MCP (talk to Codex/Gemini from Claude)

XIV. Karpathy / autoresearch¶

XV. Cognitive architecture (Etz Chaim lineage)¶

XVI. Etz Chaim AI itself¶

GitHub repo
Installation: see README.md (in repo root)
Architecture: see memory/ARCHITECTURE.md (in repo root)
Decisions: see memory/DECISIONS.md (in repo root)
Anti-patterns: see memory/MISTAKES.md (in repo root)
Portability: see docs/PORTABILITY.md → see PORTABILITY.md
2026 features map: see docs/CODE_WITH_CLAUDE_2026.md → see CODE_WITH_CLAUDE_2026.md