agentleFS
wado-langwado-lang/wado/package-loam/AGENTS.md

wado / package-loam

AGENTS.md120 starsChanged 14 days agoMITView on GitHub

Loam compiles an ONNX graph into Wado source through Kiln. Its design is in design.md, which names the documents it builds on. The roadmap and the known gaps are in TODO.md. loam gen, loam dump, the onnxruntime oracles and the benchmarks are in tools.md. WEP: Loam is complete, and nothing edits it any more. A design that changes one of its decisions is written in this directory and names the WEP's section. The documents here sketch a design and describe…

The file

We split this AGENTS.md into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not read the code of an ML framework, runtime, exporter or kernel library.

    line 17
  2. Very specific:

    Test data and models may be copied in and read, each directory carrying its source and license, as tests/onnx/ and tests/models/ do.

    line 34

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Loam Development Guide

Loam compiles an ONNX graph into Wado source through Kiln. Its design is in design.md, which names the documents it builds on. The roadmap and the known gaps are in TODO.md. loam gen, loam dump, the onnxruntime oracles and the benchmarks are in tools.md.

WEP: Loam is complete, and nothing edits it any more. A design that changes one of its decisions is written in this directory and names the WEP's section.

The documents here sketch a design and describe what exists; they decide nothing. Correctness and speed come first.

Sources

  • Do not read the code of an ML framework, runtime, exporter or kernel library.
  • Published papers and specifications may be read: ONNX's operator definitions, the StableHLO specification.
  • ONNX's own source may not be read: the code under vendor/onnx/onnx/, its reference implementation among it, vendor/onnx/tools/ and vendor/onnx/docs/docsgen/. Its headers, .proto files, docs and tests may be read, and the tests are vendor/onnx/tests/, vendor/onnx/onnx/backend/test/ and the fuzzers in vendor/onnx/onnx/fuzz/. permissions.deny enforces this for the Read tool, and .claude/hooks/vendor-license-guard.mts for Bash, which also refuses a search of a directory holding that source: name the doc to search.
  • An interface may be read, as a specification: any .proto file, and xla/pjrt/c/pjrt_c_api.h, the PJRT plugin's. The rest of XLA may not.
  • onnxruntime's operator documentation may be read, for the operators it defines beyond ONNX, such as com.microsoft::GroupQueryAttention (docs/ContribOperators.md). Its code may not. Where the documentation leaves a behavior open, onnxruntime is run as an oracle for it.
  • Test data and models may be copied in and read, each directory carrying its source and license, as tests/onnx/ and tests/models/ do.

Testing a change

Each semantics is written once per device: Cpu, WebGpuBackend, and PjrtBackend. Generated code reaches all of them, so a change can break code it never touched. Test outward from the change, as tools.md lists, and run the whole package once, last.

CI tests the logic on every device with very small programs, which should reach as many operations and paths as they can. A real model such as Gemma 3 runs by hand, and its figures go in a measurement table.

Oracles

A case in conformance/operators_test.wado or semantics_test.wado where Loam does not match onnxruntime yet is a #[TODO] test, so the run lists where the two part.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.