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.
npx skills add tonydzi/second-brain-starter-kit --skill corpus-benchgit clone --depth 1 https://github.com/tonydzi/second-brain-starter-kitWrote 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/tonydzi/second-brain-starter-kit/corpus-bench)<a href="https://agentmods.dev/skills/tonydzi/second-brain-starter-kit/corpus-bench"><img src="https://agentmods.dev/badge/skills/tonydzi/second-brain-starter-kit/corpus-bench/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/tonydzi/second-brain-starter-kit/corpus-bench"><img src="https://agentmods.dev/badge/skills/tonydzi/second-brain-starter-kit/corpus-bench.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00074 | $0.01394 |
| Opus 5.5 | $0.00030 | $0.00558 |
| Sonnet 5.5 | $0.00015 | $0.00279 |
| Haiku 4.5 | $0.00007 | $0.00139 |
Grade A, and why
corpus-bench scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/corpus-bench — верификация чужого кода на наших корпусах
Родился 14.08.2026 из ручных прогонов GIT-S5/S9 (casdk#1202: 3 дерева x 12 227 файлов руками ~час; станком — 5 минут). Приказ Антона: «Делай станок и скилл и рутину».
Движок (единственный источник логики, дока = его docstring):
python "$USERPROFILE/.claude/scripts/git25-tools/corpus_bench.py"
Когда применять
Чужой PR/фикс/релиз читает или пишет транскрипты Claude Code, сессии, память агентов, JSONL-логи — всё, для чего у нас есть живой корпус, которого нет у автора. Наше несправедливое преимущество: 12k+ живых транскриптов на узел, 5 узлов, 3 ОС.
Ритуал (4 шага)
- Цель. Папка
corpus_bench_targets/<name>/рядом с движком:target.json(repo · pinned_ref · install · metric · direction · fetch_refs для PR-голов) +plugin.pyс одной функциейevaluate(tree_path, files) -> dict(сам добавляет tree в sys.path; возвращает СЧЁТЧИКИ, никогда содержимое файлов). Образец:corpus_bench_targets/casdk-session-reader/. - Прогон.
... run <name> --refs main_sha,pr_sha [--limit 200]— сначала с--limit(быстрая проверка плагина), потом полный. Таблица: metric · ratio · delta improved/same/WORSE. Exit 3 = есть регрессия. - Мутант-дисциплина (руками, до отправки коммента): (а) red-before — метрика обязана двигаться между деревьями не вакуумно; (б) зонд загрузки — правка реально в пути исполнения (класс «невыполнившийся мутант» из журнала 12.08); (в) в коммент — только счётчики и оффсеты, leak-scan обязателен.
- Пин. Верификация закончена →
watch: trueв target.json: недельная рутина дальше сама следит, что фикс не растерялся в main.
Watch-рутина
- Задача Task Scheduler
corpus-bench-weekly(этот узел; ночное окно §4.6). - Отчёт:
$USERPROFILE/.claude/scripts/_logs/corpus_bench_last_<host>.txt— потребитель = вахта следующего захода GIT-S5 (мастер-файл GIT-25, приказ 14.08). - ALARM (exit 3) = watch_ref ухудшился >5пп с прошлого прогона → это сырьё для issue с репро, не для паники: сначала найти коммит-виновник.
- Счётчик использования:
_logs/corpus_bench_<host>.jsonl(читается на /retro).
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.
- 2d ago Changed 09a0fd4b310e
- 15d ago Changed · -2 lines dbdf0969024c
- 17d ago Changed · +2 lines · -42 tokens per session 52774d4066d1
- 20d ago Changed · +13 lines 77d0001169a8
- 29d ago First seen · 53 lines · 116 tokens per session scan A 0771c697401d
corpus-bench is a skill published in the GitHub repository tonydzi/second-brain-starter-kit (12 stars, last pushed today), licensed MIT. It adds 74 tokens to every session and 1,394 once invoked, about $0.0003 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 2026-09-24.
Other skills, from other repositories
wiki-lint
Run a deterministic, read-only health check on an Obsidian wiki. Use for lint, vault health check, audit wiki health, find orphans, find dead links, frontmatter audit, provenance audit, or wiki audit. Reports graph, link, frontmatter, provenance-ledger, empty-section, and stale-index findings; it does not reason…
lint
A health check for the LLM Wiki, a knowledge base about large language models. It looks for missing pages, broken links, conflicting or outdated content, missing fields, incomplete topic maps, weak bias checks, and an old core-context snapshot.
wiki-mode
Read or configure the vault filing methodology and suggest destinations for planned knowledge creation under Generic, LYT, PARA, or Zettelkasten. Use for wiki mode, methodology mode, what is my vault mode, set vault mode, switch to PARA, use LYT, Zettelkasten setup, change mode, configure mode, or methodology routing.…
wiki
Initialize, adopt, and route work for a separate Obsidian knowledge vault through the portable claude-obsidian core. Use for vault setup, scaffolding, workspace selection, cross-project configuration, or choosing the correct wiki sub-skill. Triggers: /wiki, set up wiki, scaffold vault, create knowledge base, adopt…
defuddle
Plan optional Defuddle extraction and retain already available local source files through reviewed filesystem capture. Use for defuddle, clean this URL, strip page clutter, readable Markdown from a web page, or preparing a web source for later wiki ingestion. Network extraction remains unavailable until an enforcing…
obsidian-markdown
Explain, draft, or validate Obsidian Flavored Markdown syntax: properties, wikilinks, embeds, callouts, tags, comments, highlights, block references, math, and Mermaid. Use when the user explicitly requests Obsidian note formatting or syntax help, not for general Markdown or broad vault operations.