Skip to content

Releases: bearatol/ceo-ai

CEO AI 0.8.0: Windows indexing and guided Docker scans

Choose a tag to compare

@bearatol bearatol released this 15 Oct 14:06
7e868ed

CEO AI 0.8.0 adds native Windows support for builtin skill and agent security indexing, plus agent-guided preparation for the optional Docker scan.

  • Builtin scanning works on macOS, Linux and native Windows with a fixed local NTFS volume. It needs Python 3.10+, without Docker or model/API calls.
  • Windows rejects unsupported paths, reparse points, package aliases and alternate data streams. Native boundary and cache tests run on Python 3.10 and 3.12.
  • Copy a preparation request from the website or guide. The agent checks compatibility, uses the recipe matching the installed release, prepares the image and smoke-tests it before scanning the selected skill.
  • The reviewed Docker recipe remains limited to native ARM64 macOS/Linux. Setup can download components and use CPU, memory and disk. Scanning runs without paid APIs.
  • Website and documentation updated in English, Russian and Simplified Chinese.

Validation: Linux, macOS and Windows CI; package integrity and documentation links; mobile website and copy-button checks. Independent security review found no remaining introduced High/Critical issue. Reviewer restrictions were instruction-based rather than technically isolated. Findings are advisory, not a safety guarantee. The Docker recipe has not received new end-to-end validation in this release.

CEO AI 0.7.0: evidence before implementation

Choose a tag to compare

@bearatol bearatol released this 15 Oct 08:53

CEO AI now checks the existing implementation and suitable components before recommending a technical approach, including the first implementation plan. Official sources resolve material gaps; sufficient local evidence can be reused. Recommendations explain the choice and label assumptions.

  • One native plugin includes CEO, Writer and Security Scan. Docker is only required for explicit deep scans.
  • Updated installation guidance and website in English, Russian and Chinese; optimized README banner.
  • Added eight behavioral acceptance scenarios, including the Mantine password toggle case. These are specifications, not a completed model benchmark.

Validation: 214 local tests, 4 skipped; package and site checks passed; 84 release Markdown files passed local link validation. Relocated scanner runtime is byte-identical to the previous implementation. Independent editorial review completed; security checklist was performed directly, without an independently sandboxed security reviewer.

CEO AI 0.6.0 — Security indexing and optional deep scans

Choose a tag to compare

@bearatol bearatol released this 13 Oct 21:07

CEO AI now includes local security observations during skill and agent indexing. The core uses Python standard-library analysis without Docker, LLM calls or external APIs.

A separate, opt-in CEO AI Security plugin provides ceo-ai-security:security-scan. Deep local scans require Python 3.10+, POSIX, Docker and a separately prepared, reviewed immutable NVIDIA SkillSpector image. Installing CEO AI does not install this module, Docker or an image.

  • Preserve partial results and explicit coverage limits; add scoped alias and TOML handling.
  • Keep findings, capability tiers and trust separate. No findings is not proof of safety.
  • Include synthetic scanner comparisons and regression tests. No hosted service, subscription or paid API client is included.
  • Update English, Russian and Simplified Chinese documentation and website.

Validation: 201 local tests, 4 platform/runtime skips; package, public documentation and site checks passed. Native Windows scanning is unsupported; POSIX runtime tests are scoped accordingly. Independent review is instruction-constrained, not a technically isolated sandbox.

Installation · Security requirements and limits

The release commit passed all five CI configurations: Linux and Windows with Python 3.10/3.12, and macOS with Python 3.12.

CEO AI 0.5.0: memory and reusable patterns

Choose a tag to compare

@bearatol bearatol released this 12 Oct 11:54

CEO AI can now keep project lessons, reusable patterns and user preferences when you explicitly enable memory. It checks existing solutions and relevant documented practices before proposing a new design.

  • Memory starts off. Project and user scopes require separate consent; saves are announced, and records can be viewed, corrected, deleted or disabled.
  • Reusable design patterns can include color tokens, component states, responsive behavior, accessibility notes and a checked example.
  • Private SQLite storage stays outside repositories and plugin caches. Installation does not create or import memory.
  • The website header stays visible while scrolling. The orange workflow signal loops along the shared path without a repeated segment or a frozen tail.
  • English, Russian and Simplified Chinese guidance explains memory controls, privacy and limitations.

The full automated suite completed with 98 tests and 4 platform/optional skips locally. The Windows, Linux and macOS CI matrix passed for this release. Responsive layouts, sticky navigation, pause/resume and reduced motion were checked in a browser. Independent static security review passed under the disclosed instruction-only isolation exception.

Memory is saved context, not model retraining. Consent and appropriate use still depend on the host agent; behavior improvements have not been established by model benchmarks. Filesystem checks assume trusted local writers. See the memory guide for storage and deletion limits.

Memory guide (EN / RU / 中文) · Installation · Website

CEO AI 0.4.0: Codex and Claude Code plugins

Choose a tag to compare

@bearatol bearatol released this 12 Oct 10:39

CEO AI can now be installed as a skills-only plugin in Codex and Claude Code from this repository's plugin catalog. Cursor keeps the project installer.

  • One canonical skill source, a generated plugin payload, and separate Codex and Claude marketplace manifests.
  • English, Russian and Simplified Chinese installation, update, removal and migration guidance.
  • No registered agents, hooks, MCP servers or added permissions.
  • Package boundaries, version consistency, cache relocation and installer ownership checks.

Native install, update, removal and skill discovery were checked on macOS with Codex CLI 0.147.0 and Claude Code 2.1.232. Model invocation remains unverified: Claude was unauthenticated, and the Codex provider rejected the configured model before execution. Windows/Linux native plugin managers and desktop UI installation flows remain unverified; the cross-platform CI covers package and project-installer behavior.

Installation and migration guide · Website

CEO AI 0.3.0: Windows installation and profession examples

Choose a tag to compare

@bearatol bearatol released this 12 Oct 07:04

CEO AI 0.3.0 adds native Windows project installation and a clearer website in English, Russian and Simplified Chinese.

  • Install through one message to your Codex, Claude Code or Cursor agent. The installer checks paths, preserves local edits and supports recovery after interruption.
  • Larger text, a pausable CEO workflow animation, continuous link underlines and six profession examples: development, marketing, sales, management, design and research.
  • Illustrated website links in the documentation, plus Telegram and LinkedIn contacts.

The default file installer passed CI on Windows (Python 3.10 and 3.12), Linux and macOS. Native skill invocation and model behavior remain unverified. APM 0.31.0 was tested on macOS with CEO AI 0.2.0; this optional check was not rerun for 0.3.0.

Website and installation · Release checks · Installation instructions

Русский

В версии 0.3.0 появилась установка в Windows через одно сообщение агенту. На сайте увеличен текст, добавлены анимация с паузой и примеры для шести профессий, исправлены подчёркивания. В документации появились заметные ссылки на сайт и контакты автора.

Установка, обновление и удаление файлов проверены на Windows, Linux и macOS. Запуск навыка внутри приложений и поведение модели пока не проверены. Проверку APM для версии 0.3.0 не повторяли.

Сайт и установка

简体中文

0.3.0 支持通过一条消息让智能体在 Windows 项目中安装 CEO AI。网站采用更大的文字,增加了可暂停的工作流程动画和六类职业示例,修复了链接下划线。文档中新增了醒目的网站入口和作者联系方式。

文件安装、更新和卸载已在 Windows、Linux 和 macOS 上验证。应用内技能调用和模型行为尚未验证;未针对 0.3.0 重新测试 APM。

网站与安装

CEO AI v0.2.0

Choose a tag to compare

@bearatol bearatol released this 11 Oct 09:33

CEO AI adds a reusable decision and execution workflow to Codex, Claude Code and Cursor.

English

  • Install one self-contained ceo-ai skill by asking your agent to follow the installation instructions.
  • Microsoft APM is an optional installation route.
  • The local installer checks ownership and local edits before updates or removal, and provides recovery for interrupted updates.
  • File installation, update and removal were tested in macOS fixture projects for all three tools, including APM 0.31.0. Invocation inside the applications remains unverified. Native Windows installation is not supported; path handling needs further work.

Русский

Установка теперь одним сообщением агенту: один навык ceo-ai со всеми нужными файлами. Есть поддержка Microsoft APM, проверка конфликтов и восстановление прерванного обновления.

Установка, обновление и удаление проверены в тестовых проектах на macOS. Запуск навыка внутри приложений пока не проверен. Нативная установка на Windows не поддерживается. Для перехода со старой ручной установки используйте актуальную помощь.

简体中文

现在可以让智能体安装一个包含所需文件的 ceo-ai 技能,也支持通过 Microsoft APM 安装。安装器会检查文件归属和本地修改,并提供更新中断后的恢复流程。

已在 macOS 临时项目中验证安装、更新和卸载;尚未验证在应用中实际调用技能。本版本不支持原生 Windows 安装。从旧版手动安装迁移时,请参阅最新帮助。

Version boundary

This release uses the unchanged v0.2.0 tag at commit 5daab9f10dc55a1878134c65e3c696c153b20e0c. Documentation and CI additions made afterward are listed under Unreleased in the current changelog; they are not part of this tag.