vllm-ascend-workspace: Skill for Codex

.agents/skills/ascend-triton-kernel-validation/SKILL.md

ascend-triton-kernel-validation is a skill for Codex from maoxx241/vllm-ascend-workspace. It costs 122 tokens per session (623 once invoked), scanned A, original, MIT.

A validation workflow for one Ascend Triton GPU kernel against a trusted reference implementation. It checks defined combinations of tensor shape, data type, memory layout, strides, options, and execution modes.

In plain words
What is it for?
Use it after writing or migrating a kernel, before making performance claims, to run planned cases on an Ascend NPU, compare results, and produce a pass manifest.
Why use it?
A successful build or benchmark does not prove that a kernel is correct; this records detailed evidence and rejects missing launches or fallback computation.

Skill for Codex

Written for Codex: agents/openai.yaml present. Also seen: installed under .agents/ (shared by several agents).

This is maoxx241/vllm-ascend-workspace's own configuration. It tells Codex how to work on vllm-ascend-workspace itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything vllm-ascend-workspace configures →

Reuse

Borrowing it

Nothing to install: this file belongs to maoxx241/vllm-ascend-workspace. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/maoxx241/vllm-ascend-workspace/main/.agents/skills/ascend-triton-kernel-validation/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/maoxx241/vllm-ascend-workspace

Made for: Codex.

Wrote 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.

agentmods badge for ascend-triton-kernel-validation

README.md
[![agentmods](https://agentmods.dev/badge/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation/github.svg)](https://agentmods.dev/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation)
Your own site
<a href="https://agentmods.dev/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation"><img src="https://agentmods.dev/badge/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for ascend-triton-kernel-validation

Your own site · 80×15
<a href="https://agentmods.dev/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation"><img src="https://agentmods.dev/badge/skills/maoxx241/vllm-ascend-workspace/ascend-triton-kernel-validation.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 122 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 623 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00122 $0.00623
Opus 5 $0.00061 $0.00311
Sonnet 5 $0.00024 $0.00125
Haiku 4.5 $0.00012 $0.00062

Measured 11d ago against content hash 185ce383a3d1, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

ascend-triton-kernel-validation 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 11d ago.

The scan reads SKILL.md. This mod also ships 3 executable files (scripts/triton_validation.py, scripts/validate_triton_impl.py, tests/test_triton_validation.py), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

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.

.agents/skills/ascend-triton-kernel-validation/SKILL.md · 43 lines

How it starts

The opening of the file, as written. The whole thing — 43 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Ascend Triton Kernel Validation

Turn a candidate into explicit compile and correctness evidence. A successful process exit or benchmark run is not correctness proof.

Workflow

  1. Freeze the candidate hash, trusted reference, predeclared tolerances, target environment, and explicit case matrix.
  2. Query .agents/knowledge/ with any observed compile, runtime, or numerical signature before repeating diagnosis.
  3. Run scripts/triton_validation.py plan. It invokes validate_triton_impl.py and rejects missing Triton kernels, a ModelNew.forward path that does not launch them, or reachable PyTorch tensor computation fallback.
  4. Before remote execution, establish remote-code-parity. Run every planned case on a managed Ascend NPU; do not run torch_npu locally.
  5. Compare shape and dtype first, then NaN/Inf behavior and numeric values. Preserve raw stdout, stderr, stack, and comparison artifacts.
  6. Normalize one result per case and run record. Do not overwrite evidence.
  7. Run analyze. Only passed_cases == total_cases > 0 produces a passed manifest.
  8. Hand the passed manifest to development or optimization. Do not benchmark a failed or inconclusive candidate.

Entry points

  • scripts/validate_triton_impl.py: AST-only fallback and launch gate.
  • scripts/triton_validation.py:
    • plan: validate config and candidate, create case matrix and Run Manifest v1;
    • record: accept one normalized case result;
    • analyze: classify the full matrix and generate the report.

Read:

Rules

  • Keep compile error, runtime error, numerical mismatch, unsupported, and missing evidence distinct.
  • Never silently cast, make inputs contiguous, remove difficult shapes, or relax tolerance to pass.
  • Load masks protect readable input addresses; store masks protect writable output positions.
  • Treat tail blocks, fully masked blocks, non-power-of-two shapes, and dynamic specialization boundaries as first-class cases.
  • Keep run state under .vaws-local/ascend-triton/validation/.

Read the full file on GitHub · 43 lines

Files

What ships with it

8 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

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.

  1. 11d ago First seen · 43 lines · 122 tokens per session scan A 185ce383a3d1

Subscribe to this mod's changes

ascend-triton-kernel-validation is a skill published in the GitHub repository maoxx241/vllm-ascend-workspace (36 stars, last pushed 7d ago), licensed MIT. It adds 122 tokens to every session and 623 once invoked, about $0.0006 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.