build-harness

build-harness is a skill for Claude Code, Codex from Dogacel/auto-gpu-kernel. It costs 29 tokens per session (985 once invoked), scanned A, original, Apache-2.0.

A repository-specific harness generator for kbench, a tool that runs correctness checks and performance benchmarks. It reads the task and repository, then writes the adapters needed to run those checks without editing the target code.

In plain words
What is it for?
Use it once, before optimization, to create correctness and benchmark adapters under the repository's harness and .omp directories.
Why use it?
It removes the repeated work of deciding how to test, measure, warm up, and validate a particular repository. It also preserves the target repository's files and Git state while preparing the harness.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions AGENTS.md.

Good fit Use it once, before optimization, to create correctness and benchmark adapters under the repository's harness and .omp directories.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/dogacel/auto-gpu-kernel/build-harness
Install

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.

Any agent
npx skills add Dogacel/auto-gpu-kernel --skill build-harness
Clone the repo
git clone --depth 1 https://github.com/Dogacel/auto-gpu-kernel

Made for: Claude Code, 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 build-harness

README.md
[![agentmods](https://agentmods.dev/badge/skills/dogacel/auto-gpu-kernel/build-harness/github.svg)](https://agentmods.dev/skills/dogacel/auto-gpu-kernel/build-harness)
Your own site
<a href="https://agentmods.dev/skills/dogacel/auto-gpu-kernel/build-harness"><img src="https://agentmods.dev/badge/skills/dogacel/auto-gpu-kernel/build-harness/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 build-harness

Your own site · 80×15
<a href="https://agentmods.dev/skills/dogacel/auto-gpu-kernel/build-harness"><img src="https://agentmods.dev/badge/skills/dogacel/auto-gpu-kernel/build-harness.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 29 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 985 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.
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 $0.00029 $0.00985
Opus 5 $0.00015 $0.00492
Sonnet 5 $0.00006 $0.00197
Haiku 4.5 $0.00003 $0.00098

Measured 2d ago against content hash ee9d630d0fbc, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-14, from the pricing page.

Security

Grade A, and why

build-harness 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 2d 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.

kopt/assets/task/skills/build-harness/SKILL.md · 107 lines

How it starts

The opening of the file, as written. The whole thing — 107 lines — stays where its author put it; the contents beside it link to each section on GitHub.

build-harness

Read the verbal brief in config.toml and AGENTS.md, inspect the untouched target repository, and generate the two adapters kbench needs. Do not optimize the target yet.

Boundary

The target clone is read-only during this turn. Record its HEAD and git status --porcelain first. You may read source, tests, documentation, build files, and history, and run existing commands. Do not edit tracked files, commit, switch branches, or run tools that rewrite lockfiles. Remove untracked files you create and finish with the same HEAD and status.

Write only under harness/ and .omp/. Never access or modify auto-gpu-kernel itself.

Understand the task

Determine from the brief and repository:

  • the target code the optimizer may edit and the public behavior it must preserve;
  • the cheapest meaningful correctness check and the comprehensive correctness check;
  • the cheap measurement useful during iteration and the representative full metric;
  • the metric unit and whether lower or higher is better;
  • required setup, warmup, synchronization, seeds, caches, and noise controls.

Prefer repo-native tests and benchmarks. The adapters may wrap any local command, but they must judge the target repository passed via --repo, not a copied implementation.

Generate harness/validate.py

It must accept exactly these common arguments:

--repo PATH
--mode quick|full
--output PATH

Kbench runs both scripts with the candidate checkout as the working directory (the clone, or a temporary worktree during kbench ab) — always use --repo, never assume the clone path. It exports KBENCH_ROOT (the project) and KBENCH_HARNESS (the harness/ directory); locate fixtures via KBENCH_HARNESS, never by relative path. Each script is killed after 30 minutes in quick mode and 60 minutes in full mode. Anything the scripts write into the candidate checkout that git does not already ignore (build outputs, caches; bytecode is already disabled) is detected as a candidate change and fails the run — keep scratch under --output's directory or a temp dir.

Read the full file on GitHub · 107 lines

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. 2d ago First seen · 107 lines · 29 tokens per session scan A ee9d630d0fbc

Subscribe to this mod's changes

build-harness is a skill published in the GitHub repository Dogacel/auto-gpu-kernel (161 stars, last pushed 2d ago), licensed Apache-2.0. It adds 29 tokens to every session and 985 once invoked, about $0.0001 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 2026-09-13.

Related

Other skills, from other repositories

review

Structured PR review for pygraphistry. Input: PR number/branch (default current branch PR). Output: findings and convergence artifacts under plans/ /. Method: multi-wave, evidence-first review across spec, correctness, tests, security, code quality, DRY, concurrency, performance, architecture, operability, and…

graphistry/pygraphistry · 69 tokens

test-amplification

Methodical bug-driven test amplification for pygraphistry features. Use when hardening a feature area via user-workflow exploration, 5-Whys retrospectives, bug-taxonomy derivation, concrete test planning, implementation, and safety-gated validation.

graphistry/pygraphistry · 56 tokens

testing

Writing frontend tests for the nvcf-ui app — Vitest, Testing Library, MSW, and project-specific render helpers. Use when creating test files, writing component tests, setting up MSW handlers in tests, deciding which render helper to use, or when the user is adding or modifying a UI component and hasn't mentioned tests…

NVIDIA/nvcf · 89 tokens

cuda-kernel-optimizer

Use when optimizing or diagnosing GPU workload performance with CUDA, CUTLASS, Triton, PyTorch, vLLM or TensorRT-LLM; interpreting NCU, Nsys or PyTorch Profiler evidence; or preparing the workload and correctness checks for such optimization. General coding, skill maintenance and conceptual GPU questions do not…

troycheng/cuda-kernel-optimizer · 77 tokens

research-engineer

An uncompromising Academic Research Engineer. Operates with absolute scientific rigor, objective criticism, and zero flair. Focuses on theoretical correctness, formal verification, and optimal implementation across any required technology.

davila7/claude-code-templates · 43 tokens

tika-eval-compare

Compare extracts from two Tika builds over a corpus to detect regressions in content, encoding, exceptions, and embedded-document handling. Use for "compare before/after extracts", "eval this change against the corpus".

apache/tika · 50 tokens