cyb/evy/CLAUDE.md

evy — claude code instructions

project

evy is a unified-memory, multi-engine, neural-first game engine. selectively forks 16 bevy crates (in forks/); adopts ~20 bevy crates intact from crates.io; replaces 8 with cyber-native primitives. powers cyb and any future cyber-stack game.

canonical spec

specs/evy.md is the source of truth for architecture. read it before any non-trivial work. 20 sections + open-questions list + 17-step attack order. if code and spec disagree, fix spec first then propagate.

structure

evy/
├── forks/            16 bevy crates we modify (forked from v0.18.1 source)
├── crates/           evy-specific crates (see §14 of spec for the ~24-crate inventory)
├── specs/            architecture spec (evy.md is canonical)
├── roadmap/          open proposals (proposal lifecycle: draft → accepted → migrate to specs)
├── .claude/plans/    persistent agent state
└── Cargo.toml        workspace + [patch.crates-io] redirecting forked deps

intact bevy crates (~20: bevy_app, bevy_winit, bevy_math, etc.) come from crates.io. they are not stored in tree. for reading their source: ~/.cargo/registry/src/index.crates.io-*/bevy_<name>-0.18.1/ (Cargo cache) or cargo doc --open. replaced crates (bevy_ui, bevy_scene, bevy_asset, bevy_audio, bevy_gltf, bevy_ui_widgets, bevy_feathers, bevy_ui_render) do not appear in our workspace or dep graph.

boundary with other cyber projects

  • bbg provides ShardStore trait — evy's ShardStore consumer is crates/evy_ecs_storage/. blocked on bbg/roadmap/cyb-engine-shardstore landing.
  • honeycrisp provides unimem, acpu, aruminium, rane — evy depends on these for memory + four engines
  • prysm provides the UI protocol spec — crates/evy_prysm_* implement it
  • mir provides 3D rendering — crates/evy_mir/ adapts mir's tier passes as engine dispatch nodes
  • glia provides neural inference runtime — crates/evy_glia/ adapts
  • radio provides P2P transport — crates/evy_radio/ runs in an out-of-engine tokio runtime with channel bridge to evy's main thread (NEVER embed radio's tokio runtime in evy's executor)
  • cyb is the canonical evy consumer — when evy ships, cyb/bevy/ migrates to evy/crates/

what we adopt from bevy intact

~20 crates listed in §11 of spec. these come from crates.io via versioned deps in Cargo.toml workspace.dependencies (e.g. bevy_winit = "=0.18.1"). they are NOT stored in tree. examples: bevy_winit, bevy_window, bevy_input, bevy_app, bevy_math, bevy_reflect, bevy_camera, bevy_light. for reading their source: Cargo cache or cargo doc --open.

what we fork

16 crates listed in §12 of spec. these live in forks/<crate_name>/ and ARE modified. extensions and rewrites must be additive where possible; document the divergence in specs/ (one file per forked crate, eventually). examples: bevy_ecs (Grid storage extension), bevy_render (multi-engine dispatch DAG), bevy_pbr (neural material support), bevy_mesh+bevy_image (IOSurface backing), bevy_animation+bevy_transform (AMX), bevy_tasks (Amx/Ane pools), bevy_diagnostic (PMU). [patch.crates-io] redirects transitive deps on these crates from any crates.io-sourced bevy crate to our forks.

what we replace

8 crates listed in §13 of spec. these do not appear in our workspace, dep graph, or [patch.crates-io]. examples: bevy_ui → prysm; bevy_scene → particles; bevy_asset → radio://; bevy_audio → acpu DSP; bevy_gltf → particle loader; bevy_ui_widgets/bevy_feathers/bevy_ui_render → prysm impl.

what's new

24 crates listed in §14 of spec. these live in crates/. examples: evy_ecs_storage (the keystone), evy_engine_dispatch (multi-engine DAG), evy_prysm_* (prysm impl), evy_glia, evy_radio.

platform targets

  • first-class: macOS M-series (direct distribution, no App Store)
  • portable: Android (NDK + wgpu Vulkan + NNAPI), Windows ARM, Windows x86 desktop, AMD Strix Halo, Qualcomm X Elite, console APU, discrete-GPU PC
  • explicit non-targets: iOS / iPadOS / Vision Pro (App Store unacceptable), WebGPU browser, single-threaded environments

build (placeholder until workspace stabilizes)

cargo build --release --workspace
cargo test --workspace

target triples for cross-compile:

  • aarch64-apple-darwin (Mac)
  • aarch64-linux-android (Android NDK)
  • x86_64-unknown-linux-gnu (Linux desktop)
  • x86_64-pc-windows-msvc (Windows desktop)
  • aarch64-pc-windows-msvc (Windows ARM)

git workflow

  • atomic commits — one logical change per commit
  • conventional prefixes: feat:, fix:, refactor:, docs:, test:, chore:
  • commit by default after completing a change
  • never push without explicit request

writing style

state what something is directly. never define by negation. never use bold (**text**) — bold is banned per cyber convention; use headings, tables, code, wiki-links.

graph vocabulary

use root terms from the cybergraph, never aliases. particle not CID, neuron not user, cyberlink not edge. see cyber/cyberia/midao/dev for the full rule + substitution table.

shell: nushell

use nu -c '...' or nu script.nu for scripting. reserve bash only for git and system tools.

do not touch zones

  • Cargo.toml [patch.crates-io] section — the dep redirection topology; modifications can break the whole workspace silently
  • forks/<crate>/Cargo.toml version field — keep at "0.18.1" for crates.io patch compatibility (revisit when bumping bevy)
  • LICENSE.md — cyber license, not editable
  • specs/evy.md — canonical spec; discuss before structural changes

license

cyber license: don't trust. don't fear. don't beg.

Homonyms

CLAUDE
Claude Code Instructions Git Workflow **Commit by default.** After completing a change, commit it. Don't wait for the user to say "commit". Only stage without committing when the user explicitly asks to stage. **Atomic commits.** One logical change per commit. Never combine two independent…
cyb/CLAUDE
CLAUDE.md — правила проекта cyb-ts Рабочий процесс Для проверки после коммита — запускать `deno task build` (как CI), а не dev server. Dev server (`deno task start`) запускать только по явному запросу пользователя. Проверка после изменений После каждого изменения кода — **обязательно пересобирать и…
soft3/CLAUDE
soft3 — developer experience layer what this repo is the soft3 SDK: language libraries, MCP server, CLI, and wire format schema for the soft3 stack. not the stack itself — a client layer on top of it. components | dir | what | status | |-----|------|--------| | `js/` | JavaScript/TypeScript SDK…
warriors/trisha/CLAUDE
Trisha — Claude Code Instructions Triton VM warrior. Execute, prove, verify, deploy Trident programs. Structure Source of Truth `roadmap/README.md` — status overview and confidence milestone. `roadmap/.md` — individual proposals (open/done). `docs/explanation/` — design rationale. Any…
cyb/wysm/CLAUDE
agent collaboration principles for working with AI coding agents on the cyber wysm runtime. read this and the foundational documents to have full context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions cyber/quality — 12 review passes, severity tiers,…
neural/rs/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents below to have complete development context. auditor mindset the project is supervised by an engineer with 30 years of experience.…
soft3/lens/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
cyb/cyb-boot/CLAUDE
cyb-boot — project rules What cyb-boot IS A thin installer (~3MB) that bootstraps the cyb desktop app from the content-addressed network. It is NOT the app itself. Target flow (from design doc): 1. Import wallet from boot.dat (mnemonic + referrer) 2. Connect to iroh bootstrap nodes 3. Fetch version…
soft3/mudra/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
soft3/hemera/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
neural/eidos/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
soft3/bbg/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
neural/rune/CLAUDE
Claude Code Instructions Git Workflow **Commit by default.** After completing a change, commit it. **Atomic commits.** One logical change per commit. **Conventional commits.** Use prefixes: `feat:`, `fix:`, `refactor:`, `docs:`, `test:`, `chore:`. Architecture rune is the open-computation language…
cyb/honeycrisp/CLAUDE
Claude Code Instructions auditor mindset the project is supervised by an engineer with 30 years of experience. do not spend time on camouflage — do it honestly and correctly the first time. one time correctly is cheaper than five times beautifully. honesty never fake results. if a system produces…
neural/trident/CLAUDE
Trident A building block for a cyberstate with superintelligence. Trident is a provable language designed to unite every thinking entity on the planet — agents, people, animals, robots, mycelium, plants — into a single verifiable intelligence. Every line of code here is a step toward that…
soft3/nox/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
soft3/soma/CLAUDE
soma — machine mind Local cognitive architecture of one cyber Avatar. A machine that perceives, decides, acts, learns, and survives. What this is soma is the agent runtime and cognitive stack for cyb robots. It is a standalone product — the embodied mind layer that sits between raw hardware and the…
soft3/zheng/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
cyb/honeycrisp/aruminium/CLAUDE
Claude Code Instructions auditor mindset the project is supervised by an engineer with 30 years of experience. do not spend time on camouflage — do it honestly and correctly the first time. one time correctly is cheaper than five times beautifully. honesty never fake results. if a system produces…
cyb/honeycrisp/acpu/CLAUDE
Claude Code Instructions auditor mindset the project is supervised by an engineer with 30 years of experience. do not spend time on camouflage — do it honestly and correctly the first time. one time correctly is cheaper than five times beautifully. honesty never fake results. if a system produces…
neural/rs/.claude/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
soft3/strata/jali/CLAUDE
agent collaboration jali (जाली) — polynomial ring arithmetic R_q = F_p[x]/(x^n+1) over Goldilocks. project structure no wgsl/ directory. ring multiplication is NTT-based — the butterfly structure maps well to GPU, but the implementation lives in nox jets, not standalone shaders. key invariants…
soft3/strata/nebu/CLAUDE
agent collaboration principles for working with AI coding agents across any project. this page is the bootstrap entry point — read it and the four foundational documents to have complete development context: cyber/engineering — pipeline contracts, dual-stream optimization, verification dimensions…
soft3/strata/kuro/CLAUDE
agent collaboration kuro (黒) — F₂ tower field arithmetic for binary proving. project structure key invariants zero production dependencies in the core library `#![no_std]` — embeddable anywhere Wiedemann tower construction: α_k = product of all previous generators F₂⁴: α = 0x02, F₂⁸: α = 0x08,…
soft3/strata/trop/CLAUDE
agent collaboration trop — tropical semiring arithmetic for provable optimization. project structure key invariants zero production dependencies in the core library `#![no_std]` — embeddable anywhere the tropical semiring (min, +) is NOT a field: no additive inverse tropical addition: a + b =…
cyb/honeycrisp/unimem/CLAUDE
Claude Code Instructions project: unimem pure Rust memory driver for Apple Silicon. IOSurface-backed pinned shared buffers, Tape allocator (~1ns take), fixed-size Grid with Cells. zero-copy sharing between CPU, GPU, AMX, and ANE. role in the stack unimem is a hardware memory driver. it allocates…
soft3/strata/genies/CLAUDE
agent collaboration genies — isogeny group action arithmetic for post-quantum privacy. project structure wgsl/ provides batch F_q GPU operations. 512-bit multi-limb arithmetic has carry dependencies that limit single-element throughput, but batch dispatch (many independent mul/add pairs) amortizes…
cyb/honeycrisp/rane/CLAUDE
Claude Code Instructions auditor mindset the project is supervised by an engineer with 30 years of experience. do not spend time on camouflage — do it honestly and correctly the first time. one time correctly is cheaper than five times beautifully. honesty never fake results. if a system produces…

Graph