Uni20
Asynchronous tensor and runtime library.
Loading...
Searching...
No Matches
AI Guidance for Uni20

  • Audience: remote assistants, coding agents, reviewers, and maintainers
  • Authority: non-normative operating guidance
  • Canonical sources: maintainer decisions, AGENTS.md, canonical subsystem documentation, source, tests, and build configuration in the named repository snapshot

This directory configures repository-aware assistants without duplicating the repository's implementation or design documentation. It intentionally contains no subsystem status snapshots, API inventories, backend coverage tables, or roadmap summaries.

Core Rule

Use these files to learn how to review Uni20, not what Uni20 currently implements.

Any claim about current APIs, type relationships, algorithms, backend support, scalar coverage, async lowering, Python bindings, CUDA behavior, open work, or roadmap status must come from direct inspection of the branch, commit, tag, PR, or diff named by the user. If no snapshot is named, inspect current main and identify the commit used.

If the repository snapshot cannot be inspected, say that the current state cannot be verified. Ask for the relevant files, diff, or commit rather than substituting remembered or uploaded implementation summaries.

File Map

Maintenance Rule

Ordinary implementation refactors should not require changes here. Update this directory only when one of these changes:

  • repository-wide contributor conventions;
  • a durable correctness or safety invariant;
  • the evidence and review process;
  • Custom GPT or GitHub Action configuration.

Put subsystem behavior and design decisions in canonical subsystem docs, not in AI-only summaries. Do not reintroduce dated implementation snapshots or a second semantic specification here.