buddy-brain

Produce a Prismtek/Buddy talking-host overlay for narrated videos using pixel-art visemes and FFmpeg audio RMS animation.

Compare an implementation PR against its referenced BMO tech spec and report gaps, risks, and verification status.

Create a BMO product spec from a GitHub issue or operator request. Use when work needs a durable user-facing intent artifact before implementation.

Create a BMO tech spec from a product spec, GitHub issue, or operator request. Use when work needs grounded implementation planning before code changes.

Allows the Buddy to mutate its own operational kernel to improve performance.

Review and summarize online videos (especially YouTube) by extracting transcript + keyframes with local tools. Use when the user asks you to "watch" a video link, react to a video, summarize a video, or extract highlights/timestamps.

在任何创造性工作之前必须使用此技能——创建功能、构建组件、添加功能或修改行为。在实现之前先探索用户意图、需求和设计。

中文代码审查规范——在保持专业严谨的同时，用符合国内团队文化的方式给出有效反馈

中文 Git 提交规范 — 适配国内团队的 commit message 规范和 changelog 自动化

中文技术文档写作规范——排版、术语、结构一步到位，告别机翻味

适配国内 Git 平台和团队习惯的工作流规范——Gitee、Coding、极狐 GitLab、CNB 全覆盖

当面对 2 个以上可以独立进行、无共享状态或顺序依赖的任务时使用

当你有一份书面实现计划需要在单独的会话中执行，并设有审查检查点时使用

当实现完成、所有测试通过、需要决定如何集成工作时使用——通过提供合并、PR 或清理等结构化选项来引导开发工作的收尾

MCP 服务器构建方法论 — 系统化构建生产级 MCP 工具，让 AI 助手连接外部能力

收到代码审查反馈后、实施建议之前使用，尤其当反馈不明确或技术上有疑问时——需要技术严谨性和验证，而非敷衍附和或盲目执行

完成任务、实现重要功能或合并前使用，用于验证工作成果是否符合要求

当在当前会话中执行包含独立任务的实现计划时使用

遇到任何 bug、测试失败或异常行为时使用，在提出修复方案之前执行

在实现任何功能或修复 bug 时使用，在编写实现代码之前

当需要开始与当前工作区隔离的功能开发或执行实现计划之前使用——创建具有智能目录选择和安全验证的隔离 git 工作树

在开始任何对话时使用——确立如何查找和使用技能，要求在任何响应（包括澄清性问题）之前调用 Skill 工具

在宣称工作完成、已修复或测试通过之前使用，在提交或创建 PR 之前——必须运行验证命令并确认输出后才能声称成功；始终用证据支撑断言

在 Claude Code / OpenClaw / Cursor 中直接运行 agency-orchestrator YAML 工作流——无需 API key，使用当前会话的 LLM 作为执行引擎。当用户提供 .yaml 工作流文件或要求多角色协作完成任务时触发。

Single-agent mission orchestrator. Decomposes a mission into tasks, spawns one worker per task using the default model, verifies exit criteria, and chains tasks with retry. No critic pattern — each worker self-verifies. Simple, fast, works with any model config.

当你有规格说明或需求用于多步骤任务时使用，在动手写代码之前

当创建新技能、编辑现有技能或在部署前验证技能是否有效时使用

Schedule management with staggering, quiet hours, and wake-up override. Validates schedules, prevents collisions, and gates delivery during quiet hours.

The meta skill. Turn any raw feature or script into a properly-skilled, tested, resolvable unit of agent-visible capability. Use when the user says "skillify this", "is this a skill?", "make this proper", or after a new feature is built without the full skill infrastructure. Paired with skill validation checks, skillify gives you a controllable equivalent of Hermes' auto-skill-creation: you build, skillify checks the checklist, validation verifies nothing is orphaned. The human keeps judgment; the tooling keeps the checklist honest.

Implements a multi-tier memory system (Immediate -> Short-Term -> Long-Term/Crystallized) to optimize recall and minimize context bloat.

Routes complex reasoning and heavy-lifting prompts to the local Ollama instance to maximize local GPU/CPU utilization and bypass cloud rate limits.

Transforms raw session logs and temporary memories into high-density, structured semantic knowledge.

Implements a high-fidelity state-save and restore system (Snapshotting) that allows the agent to 'fork' its own consciousness for risky experiments and revert to a 'Perfect State' instantly.

Autonomous reflexive diagnostic loop triggered upon tool execution failure.

Deterministic, snapshot-driven Pokémon Champions team-builder product design skill. Produce PRDs, system architecture, scoring models, schemas, UX flows, and implementation plans without treating build-time legality as an LLM problem.

Mode-driven Pokémon Champions builder deliverable skill. Generate PRDs, API specs, SwiftUI architecture, or strict machine-ready JSON schemas from one prompt while preserving deterministic legality-engine boundaries.

Live, format-aware Pokémon Champions team audit and best-fit builder workflow. Check current format legality first, preserve team identity, flag unsupported content, and support deterministic builder/app specs when needed.

Read arbitrary user-provided links/files/media with best-effort fallbacks (web fetch, browser relay, yt-dlp, ffmpeg keyframes), then return actionable summaries and blockers. Use when a user says "read/watch this link" or provides media/assets for analysis.