Borrowing it
Nothing to install: this file belongs to albumentations-team/albucore. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/albumentations-team/albucore/main/.codex/skills/albucore-benchmarks/SKILL.mdgit clone --depth 1 https://github.com/albumentations-team/albucoreWrote 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.
[](https://agentmods.dev/skills/albumentations-team/albucore/albucore-benchmarks)<a href="https://agentmods.dev/skills/albumentations-team/albucore/albucore-benchmarks"><img src="https://agentmods.dev/badge/skills/albumentations-team/albucore/albucore-benchmarks/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.
<a href="https://agentmods.dev/skills/albumentations-team/albucore/albucore-benchmarks"><img src="https://agentmods.dev/badge/skills/albumentations-team/albucore/albucore-benchmarks.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00048 | $0.02219 |
| Opus 5.5 | $0.00019 | $0.00888 |
| Sonnet 5.5 | $0.00010 | $0.00444 |
| Haiku 4.5 | $0.00005 | $0.00222 |
Grade A, and why
albucore-benchmarks 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 yesterday.
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.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Albucore Benchmarks
Albucore benchmark scope
Measure Albucore public operations only, including Core-owned dispatch, conversions, allocations and memory costs. AX functional, transform, Compose and DataLoader performance campaigns require an explicit user request and never become automatic acceptance gates for Core. Choose a matrix for the affected operations and routing boundaries; the canonical guide below defines the detailed scope.
Before designing a performance comparison, read ../performance-optimization/SKILL.md and
../../../docs/performance-optimization.md completely. Extend the benchmark along the dimension that controls the
candidate, such as label density for bincount, table and channel layout for LUTs, or output size and dtype for random
generation.
Use Python 3.11 and the latest stable third-party releases compatible with Python 3.11 and the target platform for
acceptance benchmarks. The recorded 2026-10-31 profile includes PyTorch 2.15.0 (2.15.0+cpu on Linux), NumPy 2.4.6,
OpenCV 5.0.0.93, NumKong 7.8.6, and StringZilla 5.3.2. Re-resolve and record all package versions and imported source
paths before each acceptance run. uv run follows uv.lock; verify its resolved profile before using results as
acceptance evidence.
Use exactly one CPU thread per process for every candidate, following the thread controls in the canonical
performance guide. Benchmark additional thread counts only when the user explicitly requests thread scaling.
Record enabled backend/ISA flags and verify the intended host profile before acceptance. A sandbox that denies
sysctl can disable NumKong NEON. On macOS ARM64, invoke local benchmark commands through Codex with
sandbox_permissions="require_escalated" and verify nk.get_capabilities()["neon"] is true in the same process. The
router harness rejects this restricted profile by default; --allow-restricted-isa is for diagnostics only. Do not
force NumKong capabilities to make an acceptance run appear native.
For dtype comparisons, use the same logical inputs and operation controls; choose floating controls exactly
representable in float16 when the route is compared across float16 and float32. Do not benchmark an identity route for
one dtype against real work in another.
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.
- yesterday Changed · +14 lines 400fcf2eb639
- 4d ago Changed · +19 lines 7876ca9ca37e
- 24d ago Changed · +5 lines 5946ec5e0549
- 1mo ago Changed · +8 lines 4bfe0b04d9d0
- 2mo ago Changed · +3 lines d4013c6acc7f
- 2mo ago First seen · 84 lines · 48 tokens per session scan A 6207800e02b0
albucore-benchmarks is a skill published in the GitHub repository albumentations-team/albucore (124 stars, last pushed yesterday), licensed MIT. It adds 48 tokens to every session and 2,219 once invoked, about $0.0002 per session on Opus 5.5. 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.
albucore-benchmarks compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| albucore-benchmarksalbumentations-team/albucore | Skill | 124 | yesterday | MIT | A | 48 | Codex |
| lepton-clileptonai/leptonai | Skill | 2.8k | yesterday | Apache-2.0 | A | 122 | Claude Code |
| Computer Vision Helpereddiebelaval/squire | Skill | 21 | 2mo ago | MIT | A | 16 | Claude Code, Codex |
| platform-cliultralytics/skills | Skill | 35 | yesterday | AGPL-3.0 | A | 159 | Claude Code, Codex |
| ml-for-aecAbhinavbwj/Claude-skills-for-Computational-Designers | Skill | 226 | 7mo ago | MIT | A | 39 | Claude Code |
| yolo-exportultralytics/skills | Skill | 35 | yesterday | AGPL-3.0 | A | 120 | Claude Code, Codex |
| yolo-trainingultralytics/skills | Skill | 35 | yesterday | AGPL-3.0 | A | 95 | Claude Code, Codex |