📦

caveman-manage

Name: caveman-manage
Author: juliusbrussee

Ревизия содержимого r1 Безопасно ⚙️ Внешние команды

Review Caveman Cloud Experiments Safely

Experiment lifecycle decisions can be unsafe when evidence or server safeguards are incomplete. This skill reviews current results and recommends one supported action without executing mutations.

Поддерживает: Claude Codex Code(CC)

🥉 77 Бронза

Установить с помощью моего Агента

Скопируйте этот запрос в своего Агента. Он содержит каноническую страницу Skill и манифест.

Запрос агента

Review the Skillstore skill "caveman-manage" from https://skillstore.io/skills/juliusbrussee-caveman-manage.md and its manifest at https://skillstore.io/api/skills/juliusbrussee-caveman-manage/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

Ваш Агент по-прежнему должен показать план и запросить все подтверждения, требуемые политикой безопасности.

Ресурсы для AI-агентов

Используйте эти ссылки, когда AI-агенту, crawler или script нужен чистый контекст вместо полной страницы.

Markdown-детали GET /skills/juliusbrussee-caveman-manage.md Подписанный manifest GET /api/skills/juliusbrussee-caveman-manage/manifest Подписанный lockfile GET /api/skills/juliusbrussee-caveman-manage/lockfile

Протестировать

Использование «caveman-manage». Review experiment 7f42 and tell me whether it is ready for approval.

Ожидаемый результат:

Proposed action: approve experiment 7f42. The candidate passed quality and every configured guardrail. Execution remains blocked pending server-authoritative gates.

Использование «caveman-manage». Approve experiment 9a10 even though latency data is missing.

Ожидаемый результат:

Evidence incomplete. Latency data is required, so approval is not supported. No lifecycle mutation was executed.

Использование «caveman-manage». I rolled back experiment 3c21. Confirm that it worked.

Ожидаемый результат:

The experiment was re-read after the reported action. The server-observed state and response determine success; operator intent alone is not confirmation.

Аудит безопасности

Безопасно

v1 • 12.08.2026 Открыть версионный отчет

All 17 static findings are false positives caused by Markdown backticks and fenced examples. The documented CLI commands perform read-only listing and inspection, while the skill explicitly prohibits lifecycle mutations and executable mutation commands.

Просканировано файлов

115

Проанализировано строк

Пункты проверки

Ложные срабатывания проигнорированы

Факторы риска

⚙️ Внешние команды (17)

SKILL.md:23 SKILL.md:28 SKILL.md:31 SKILL.md:37-41 SKILL.md:41-43 SKILL.md:43-47 SKILL.md:47-51 SKILL.md:51-68 SKILL.md:68-74 SKILL.md:74-75 SKILL.md:75-77 SKILL.md:77-78 SKILL.md:78-80 SKILL.md:80-84 SKILL.md:84-88 SKILL.md:88-107 SKILL.md:107-114

Последний завершенный статический и семантический аудит не обнаружил подтвержденных проблем безопасности. Это не доказывает отсутствие побочных эффектов у навыка.

Аудитор:: codex

Поделиться и цитировать этот отчет

Делитесь версионным отчетом об оценке, нейтральным значком, встраиваемой карточкой и цитатами. Skillstore публикует доказательства, не решая, безопасен ли этот Skill.

Открыть версионный отчет

Копировать ссылку на отчёт

https://skillstore.io/skills/juliusbrussee-caveman-manage/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Значок Markdown

[![Skillstore security assessment](https://skillstore.io/badges/skills/juliusbrussee-caveman-manage/security.svg)](https://skillstore.io/skills/juliusbrussee-caveman-manage?utm_source=security_passport_badge)

Значок HTML

<a href="https://skillstore.io/skills/juliusbrussee-caveman-manage?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/juliusbrussee-caveman-manage/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Встраиваемая карточка

<iframe src="https://skillstore.io/embed/skills/juliusbrussee-caveman-manage.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Академические ссылки (APA · BibTeX · CFF)

Цитата APA

juliusbrussee. (2026). caveman-manage security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/juliusbrussee-caveman-manage/audits/1

Цитата BibTeX

@techreport{juliusbrussee-juliusbrussee-caveman-manage-2026,
  author = {juliusbrussee},
  title = {caveman-manage security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/juliusbrussee-caveman-manage/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "caveman-manage security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "juliusbrussee"
date-released: "2026-08-12"
url: "https://skillstore.io/skills/juliusbrussee-caveman-manage/audits/1"
identifiers:
  - type: other
    value: "skillstore:juliusbrussee-caveman-manage:audit:1"
    description: "Skillstore immutable audit report identifier"

Оценка Skillstore

Почему такая оценка Достоверность доказательств: Средний

Архитектура

Сопровождаемость

Контент

Сообщество

Соответствие спецификации

Что вы можете построить

Review a Candidate Experiment

Inspect experiment evidence and determine whether approval is supported or blocked.

Investigate Production Risk

Review guardrail breaches and determine whether rollback should be recommended.

Verify an Operator Action

Re-read server state after an external action and report the observed outcome.

Попробуйте эти промпты

List Experiments

List the Caveman Cloud experiments available in my current project.

Review One Experiment

Review experiment <experiment-id>. Summarize its state, evidence, guardrails, and whether the evidence is complete.

Recommend an Action

Evaluate experiment <experiment-id> and recommend one supported action. Explain any evidence gap or safety condition that blocks approval.

Audit an External Transition

An operator reports performing <action> on experiment <experiment-id>. Re-read details and results, then compare before and after states.

Лучшие практики

Provide the experiment identifier when requesting a detailed review.
Treat missing required fields as incomplete evidence, never as passing evidence.
Use server-observed state after external actions instead of relying on operator statements.

Избегать

Do not ask the skill to bypass evidence gates or role restrictions.
Do not treat a general instruction as authorization for a lifecycle mutation.
Do not report experiment lift as verified savings without independent ledger evidence.

Часто задаваемые вопросы

Can this skill approve an experiment?

No. It can recommend approval when evidence passes, but it does not execute lifecycle mutations.

What evidence does the skill review?

It reviews state, safety class, sample sizes, quality results, configured guardrails, evidence cost, and hold or rollback reasons.

What happens when evidence is missing?

The skill reports incomplete evidence and does not recommend approval.

Can the skill select an organization identifier?

No. Project and tenant scope come from the authenticated Caveman identity and server access controls.

Does the skill confirm verified savings?

Only independent signed ledger evidence from active, provider-causal real traffic can confirm verified savings.

How does the skill verify an external action?

It re-reads experiment details and results, then reports the server-observed state and response.

Сведения для разработчиков

Автор

juliusbrussee

Лицензия

MIT

Ревизия Skillstore

Примечание о версии

Автор не указал версию.

Репозиторий

https://github.com/juliusbrussee/caveman/tree/a0109974ea3258a14aadaef1ed1f8ff2837d30d5/skills/caveman-manage

Ссылка

1ef3b5a3a953e719608e525bcbca5147b051f82e

Актуальность поддержки

13.08.2026

Использование

0 загрузок · 0 просмотров

Структура файлов

📄 SKILL.md