Skip to content
Agentmart
Submit a tool

cuda-kernel-debug

Narrow a CUDA kernel correctness failure to the first failing coordinate and one semantic owner using minimal probes.

Skill by linjuhon

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/linjuhon-cuda-kernel-debug

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Narrow a CUDA kernel correctness failure to the first failing coordinate and one semantic owner using minimal probes. Use only when the active gate requires CUDA failure isolation.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/cuda-kernel-debug

What's inside

Skill name
cuda-kernel-debug
Description
Narrow a CUDA kernel correctness failure to the first failing coordinate and one semantic owner using minimal probes. Use only when the active gate requires CUDA failure isolation.
License
MIT
Instructions
32 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# cuda-kernel-debug

Use this skill when the active gate requires narrowing a CUDA kernel failure.

## Responsibilities

- isolate the first failing coordinate
- keep the probe minimal
- preserve measurement identity
- separate row, lane, store, and dispatch ownership

## Required Output

- failure coordinate
- owning range
- minimal probe used
- preserved passes
- next falsifiable hypothesis

## Hard Rules

- Do not patch until the owner is unique.
- Do not expand scope while the first divergence is unresolved.
- Do not mix correctness debugging with performance work.

## Entry And Return

Read the active gate, freeze, and prior evidence before probing. Return one
minimal probe, the first divergence, candidate owner, preserved passes, and one
falsifiable next action. Do not modify source until the owner is unique.

From linjuhon/codex-personal-skills, MIT.

More from linjuhon/codex-personal-skills

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Aug 8, 2026 (2 months ago)
Last release
Aug 8, 2026 (2 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from linjuhon

Cite this page

cuda-kernel-debug skill. Agent Mart. https://agentmart.dev/skills/linjuhon-cuda-kernel-debug (last updated 2026-10-24)