benchmark

A procedure for running OfficeDocBench, a benchmark for comparing document parsers, and keeping the resulting scores synchronized across the AILANG Parse website.

In plain words
What is it for?
Use it to run a full or AILANG Parse-only evaluation, regenerate the score summary, update the website data, and check for score regressions.
Why use it?
It prevents benchmark results from becoming inconsistent between the main results file, the website's copy, and pages that contain manually maintained data.

Skill for Claude CodeCodex

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.

agentmods
npx agentmods add skills/sunholo-data/ailang-parse/benchmark
Any agent
npx skills add sunholo-data/ailang-parse --skill benchmark
Clone the repo
git clone --depth 1 https://github.com/sunholo-data/ailang-parse

Made for: Claude Code, Codex.

Per session 115 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,498 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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 $0.00115 $0.01498
Opus 5 $0.00057 $0.00749
Sonnet 5 $0.00023 $0.00300
Haiku 4.5 $0.00012 $0.00150

Measured yesterday against content hash 1375f1f5a687, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

benchmark 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.

The scan reads SKILL.md. This mod also ships 4 executable files (scripts/_postcheck.sh, scripts/full.sh, scripts/quick.sh, …), 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.

.claude/skills/benchmark/SKILL.md · 118 lines

How it starts

The opening of the file, as written. The whole thing — 118 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Benchmark

Run OfficeDocBench, regenerate summary.json (the single source of truth for benchmark numbers across the website), and verify the docs site stays in sync.

Why this skill exists

OfficeDocBench scores live in one place: benchmarks/officedocbench/results/summary.json. From there they propagate to the static site via docs/data/officedocbench-summary.json (a mirror) and the data-bench injector at docs/js/bench-data.js. If you re-run the benchmark without using this skill, you risk:

  • Forgetting to regenerate the summary mirror in docs/
  • Forgetting that JSON-LD / SEO copy on a handful of pages cannot use data-bench and must be hand-edited if the AILANG Parse composite changes
  • Not catching score regressions before committing

This skill bundles all of those steps so they happen together.

Three modes

When Use script What it does
Changed parser code, want to re-verify everything scripts/full.sh Full eval across all 8 adapters (~1 minute), regenerates summary, syncs docs
Changed parser code, only care about AILANG Parse scripts/quick.sh docparse-only eval (~10 sec), regenerates summary, syncs docs
Just want to refresh summary.json from existing per-adapter results without re-parsing scripts/refresh-summary.sh Re-aggregates the existing results/<adapter>/results.json files into a fresh summary.json + mirror

After any of these, the script automatically:

  1. Diffs the new AILANG Parse composite score against the previous run
  2. If the composite changed, prints a checklist of static-copy files that need manual updates (since they can't use data-bench)
  3. Reports the headline numbers so you can sanity-check before committing

Quick start

# Standard: re-run all adapters and refresh the site
.claude/skills/benchmark/scripts/full.sh

# Fast iteration: only re-run AILANG Parse
.claude/skills/benchmark/scripts/quick.sh

# Just regenerate summary.json from existing results (no parsing)
.claude/skills/benchmark/scripts/refresh-summary.sh

Read the full file on GitHub · 118 lines

Files

What ships with it

5 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. yesterday First seen · 118 lines · 115 tokens per session scan A 1375f1f5a687

Subscribe to this mod's changes

benchmark is a skill published in the GitHub repository sunholo-data/ailang-parse (0 stars, last pushed yesterday), licensed Apache-2.0. It adds 115 tokens to every session and 1,498 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 2026-08-31.

Related

Other skills, from other repositories

document-converter

Convert Office documents (PPTX, DOCX, XLSX, PDF, HTML, CSV, JSON, XML, images) to Markdown using Microsoft MarkItDown. Provides the agent with conversion strategies for academic and research workflows.

wentorai/Research-Claw · 50 tokens

office-router

Route Office requests into the correct xlsx, docx, or pptx workflow, while preferring Slidev for new presentation drafts unless .pptx is explicitly required.

matingai/crab · 38 tokens

doc-converter

Universal document format converter - DOCX, PDF, Markdown, HTML, TXT.

chainlesschain/chainlesschain · 19 tokens

document-processing

End-to-end document processing agent skill covering PDF manipulation (extract, merge, split, rotate, watermark, form-fill, OCR), DOCX creation and editing (templates, mail-merge, style management), XLSX spreadsheet handling (formulas, charts, pivot tables, data analysis), PPTX presentation generation (layouts, charts…

JPeetz/agent-skills · 174 tokens

docx-architecture-audit

Audit the office-open-xml-viewer DOCX layout architecture for a single immutable layout-to-paint pipeline. Use after major DOCX layout, pagination, measurement, paint, parser-model, worker, or compatibility changes, and before declaring Issue.

yukiyokotani/office-open-xml-viewer · 55 tokens

office-artifacts

Use when creating, opening, reading, or editing Office artifacts such as LibreOffice-native ODT/ODS/ODP files and compatibility DOCX/XLSX/PPTX files with the officeartifact tool.

agent0ai/agent-zero · 47 tokens