Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/lulzx/cuda-metal/agents-mdgit clone --depth 1 https://github.com/Lulzx/cuda-metalWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/lulzx/cuda-metal/agents-md)<a href="https://agentmods.dev/instructions/lulzx/cuda-metal/agents-md"><img src="https://agentmods.dev/badge/instructions/lulzx/cuda-metal/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01065 | $0.01065 |
| Opus 5 | $0.00532 | $0.00532 |
| Sonnet 5 | $0.00213 | $0.00213 |
| Haiku 4.5 | $0.00106 | $0.00106 |
Grade A, and why
cuda-metal AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — CuMetal Development Guide
This file guides day-to-day implementation work in this repository.
Start with /Users/lulzx/work/cumetal/spec.md before architectural changes.
If there is a conflict:
spec.mdwins.AGENTS.mdis next.README.mdis status-oriented and may lag.
Project Identity
CuMetal is a CUDA compiler/runtime targeting Apple Metal on Apple Silicon.
- Primary path: source recompilation with
cumetalc. - Secondary path: opt-in binary-shim compatibility (
libcuda.dylibalias and__cudaRegister*path) whenCUMETAL_ENABLE_BINARY_SHIM=ON.
Current Focus
The repository is beyond the bootstrap phases. Focus implementation on:
- Phase 4 correctness hardening (conformance, ABI compatibility, negative-path behavior).
- Phase 4.5 shim coverage quality (cuBLAS/cuRAND behavior and tests).
- Binary-shim hardening (additional fatbinary variants and registration parity) without regressing the source-first path.
- AIR ABI tooling hardening where practical (
air_inspect,air_validate), with explicit limitations tracked indocs/known-gaps.md.
Repository Structure
cumetal/
├── compiler/
│ ├── frontend/ # .cu frontend path
│ ├── ptx/ # PTX parser/lowering
│ ├── passes/ # LLVM transformation passes
│ ├── air_emitter/ # LLVM IR -> metallib emission
│ ├── air_validate/ # metallib validation
│ └── cumetalc/ # CLI
├── runtime/
│ ├── api/ # clean-room CUDA headers
│ ├── driver/ # Driver API shim
│ ├── rt/ # Runtime API shim
│ ├── registration/ # __cudaRegister* path (opt-in binary shim)
│ ├── error/ # per-thread error queue
│ ├── cache/ # metallib cache
│ └── metal_backend/ # Objective-C++ Metal boundary
├── tools/
│ ├── air_inspect/
│ ├── metal_library_archive_bridge/
│ ├── ptx_diff/
│ └── cumetal_bench/
├── tests/
│ ├── unit/
│ ├── functional/
│ ├── conformance/
│ ├── air_abi/
│ └── ptx_sweep/
├── docs/
│ ├── air-abi.md
│ └── known-gaps.md
├── spec.md # canonical design spec
└── AGENTS.md # this file
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 122 lines · 1,065 tokens per session scan A e832cb53980c
cuda-metal AGENTS.md is an instructions file published in the GitHub repository Lulzx/cuda-metal (167 stars, last pushed 3d ago), licensed Apache-2.0. It adds 1,065 tokens to every session, about $0.0053 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
apfel CLAUDE.md
Instructions for Arthur-Ficial/apfel, covering apfel - project instructions, the golden goal, core product (this is what apfel is), byproducts (useful, but not the pitch) and readme.md structure rule.
flashinfer AGENTS.md
AGENTS.md instructions for flashinfer-ai/flashinfer: This repository’s agent instructions live in CLAUDE.md.
speech-swift AGENTS.md
AGENTS.md instructions for soniqo/speech-swift, covering agent instructions, communication style, workflow, running tests and benchmarks — sequential and memory-aware and git conventions.
cuopt AGENTS.md
AGENTS.md instructions for NVIDIA/cuopt, covering agents.md — cuopt ai agent entry point, skills directory (flat), rules, common (concepts only; no api code) and installation.
Vocello AGENTS.md
AGENTS.md instructions for PowerBeef/Vocello, covering agents.md — vocello (qwenvoice), product and authority, start and resume work, hard invariants and domain routing.
humla CLAUDE.md
Claude Code instructions for michaelwilhelmsen/humla, covering humla — project notes, what this app is, core capabilities, architecture overview and data flow during a recording.