AI & ML

ngxtm-devkit Skill

Use when working with performance testing review multi agent review

rysweet Bundle

Iterative codebase quality audit with multi-agent validation and escalating-depth SEEK/VALIDATE/FIX/RECURSE cycle. Enforces fix-all-per-cycle rule — every confirmed finding must be fixed before moving to the next cycle. Use for quality audit, code audit, codebase review, technical debt audit, refactoring opportunities, module quality check, or architecture review.

rysweet Skill

TLA+ formal specification expert for writing specs, model checking, and applying formal methods to amplihack workflows

rysweet Bundle

Comprehensive knowledge of Claude Agent SDK architecture, tools, hooks, skills, and production patterns. Auto-activates for agent building, SDK integration, tool design, and MCP server tasks.

rysweet Skill

Autonomous co-pilot — agent formulates goal from natural language, enables lock mode with SessionCopilot reasoning, works until goal is achieved.

rysweet Bundle

Extracts key learnings from conversations, debugging sessions, and failed attempts. Use at session end or after solving complex problems to capture insights. Stores discoveries in memory (via amplihack.memory.discoveries), suggests PATTERNS.md updates, and recommends new agent creation. Ensures knowledge persists across sessions via Kuzu memory backend.

rysweet Bundle

Microsoft Agent Framework Skill

rysweet Bundle

Guides architects on when and how to use goal-seeking agents as a design pattern. This skill helps evaluate whether autonomous agents are appropriate for a given problem, how to structure their objectives, integrate with goal_agent_generator, and reference real amplihack examples like AKS SRE automation, CI diagnostics, pre-commit workflows, and fix-agent pattern matching.

rysweet Bundle

Production-ready skill for detecting silent degradation across codebases. Uses multi-wave audit system with 6 specialized category agents, multi-agent validation panel, and convergence detection.

rysweet Skill

Encodes a continuous improvement loop for goal-seeking agents: EVAL, ANALYZE, RESEARCH (hypothesis + evidence + counter-arguments), IMPROVE, RE-EVAL, DECIDE. Auto-commits improvements (+2% net, no regression >5%) and reverts failures. Works with all 4 SDK implementations. Auto-activates on "improve agent", "self-improving loop", "agent eval loop", "benchmark agents", "run improvement cycle".

rysweet Bundle

Generates comprehensive end-to-end test scenarios using outside-in methodology. Supports Web (Playwright), CLI, TUI, API, and MCP/gadugi-style scenarios. Auto-detects app type or accepts explicit override.

hironow Bundle

Interactive QA session where user reports bugs or issues conversationally, and the agent files GitHub issues. Explores the codebase in the background for context and domain language. Use when user wants to report bugs, do QA, file issues conversationally, or mentions "QA session".

hironow Bundle

Exploratory testing of a running web application through a real browser (agent-browser): systematically drive the UI, find functional/UX/console issues, and produce a report with screenshot-and-video repro evidence per issue. Use for "dogfood", "bug hunt", or "exploratory test this site/app"; for filing issues the user reports conversationally, use the qa skill.

hironow Bundle

Tell the agent to zoom out and give broader context or a higher-level perspective. Use when you're unfamiliar with a section of code or need to understand how it fits into the bigger picture.

hironow Bundle

Use when the user asks how to build with OpenAI products or APIs and needs up-to-date official documentation with citations (for example: Codex, Responses API, Chat Completions, Apps SDK, Agents SDK, Realtime, model capabilities or limits); prioritize OpenAI docs MCP tools and restrict any fallback browsing to official OpenAI domains.

hironow Bundle

Run very long-horizon, multi-hour autonomous builds in Claude Code by splitting roles - a manager session that interviews the user, writes a phased checklist and steers, and a separate implementer agent that finishes one phase at a time "extremely well" while ticking a progress dashboard. Use whenever the user asks for an ambitious end-to-end build, a large migration or rewrite, a "build the whole thing" request, or anything a human team would need days for; also when the user says "manager loop", "run this to completion", "keep going until it's done", "long-running", "autonomous build", or complains that a long agent run stalled in minutiae, plateaued, or asymptoted. Not for ordinary single-task coding.

hironow Skill

Audit AGENTS.md / CLAUDE.md / GEMINI.md and their spoke files (docs/agents/* etc.), classify every line as discoverable / task-specific / always, and produce a staged plan to shrink the hub and extract skills. Use whenever the user wants agent instruction files audited, slimmed, reorganized, or skill-ified, or asks about context/token reduction or an instruction budget for those files — whether or not they say "audit".

maslennikov-anton Bundle

Validate VCont Megapack .run artifacts and inspect /home/ant/IdeaProjects/megapack: release-build CI, manifest versions, x86_64/aarch64 matrix, Nexus paths, opcua/agent flags, install/uninstall/purge risks.

maslennikov-anton Bundle

Управлять контекстом текущей agent-сессии: compaction/resume, handoff, state snapshot и снижение токенового шума; not persistent memory writes.

maslennikov-anton Bundle

Проектировать и проверять security: auth/authz, secrets, dependency risks, attack surface, secure defaults, threat model и controls.

maslennikov-anton Bundle

Создавать или менять Paperclip agent adapter: пакет адаптера, интерфейсы, модули, регистрацию и поддержку CLI/API/custom AI coding tools.

maslennikov-anton Bundle

Использовать, когда пользователь разрешил subagents, parallel agents, delegation или agent workflow, и задача содержит независимые subtasks для крупной реализации.

2799662352 Skill

Optimizes user prompts for Seedance 2.0 multi-modal video generation. Invoke when a user provides an initial prompt or multimedia assets for video generation, or explicitly requests prompt optimization.

2799662352 Bundle

Process video/audio through the ffmpeg-win MCP tool (Dockerized FFmpeg 8.1 with automatic Windows-path conversion) instead of a local ffmpeg binary. Use for transcoding, resizing, trimming, speed change, compression, audio extraction, concat, cropping, fades, overlays, thumbnails, GIFs, and inspection — every command is one ffmpeg-win call with an args array plus a drive-root basedir. Triggers on "用 ffmpeg", "处理视频", "转码/压缩/裁剪/拼接视频", "提取音频", "竖屏适配", "加 BGM", "ffmpeg-win", or any CATIMATION 出片 post-processing. References cover filters, codecs, audio, streaming/hwaccel, platform export, and the CATIMATION workflow.

2799662352 Bundle

アニメ映像(動画)を作る作画工芸スキル。アニメ短編・絵コンテ・アニマティック・AI動画ショット・キャラ芝居・カット尺・カメラワーク・アクションのビート・連続性・編集リズム・効果音の手がかり、または Seedance/即夢/可灵/海螺 向けの画像から動画(i2v)ショットprompt を作成・推敲・レビューするときに使う。中核は日本のアニメ制作の作画技法——タメ・ツメ、緩急、メリハリ、間、コマ打ち(1/2/3コマ)、動勢、慣性、フォロースルー、予備動作、残像・ブレ・効果線、シルエット、質量感、レイアウト、撮影——で、画風層(director-anime-quality-boost)が扱わない「運動の法則層」を補う。Professional Japanese animation (sakuga) craft skill for animated shorts, e-konte, character acting, and image-to-video shot prompts. 触发词/トリガー:アニメ、作画、動画、絵コンテ、アニマティック、中割り、タメツメ、緩急、メリハリ、間、コマ打ち、動勢、残像、効果線、レイアウト、キャラ芝居、i2v、动画、动画短片、动画分镜、动画12原理、运动法则、角色表演、图生视频、卡通、animation、animated short、sakuga、motion principles、character acting。注意:UI/フロントエンドのモーションには使わない(ユーザーが明示的にインターフェースアニメを求めた場合を除く)。

2799662352 Skill

【分镜模式·音频 / Storyboard · Audio】触发词:配乐 / 音效 / 配音 / 声音设计 / BGM / 三层音频 / 乐器 / bpm / 混响 / score / SFX / voice / sound design。Use when each shot needs a three-layer audio field — A1 Score (ref:Composer/Work + 乐器 + 力度 pp-ff + 速度 bpm + 张力值 0-10, derived via score_bpm = motion_freq × 60), A2 SFX (材质 + 动作 + 频率Hz + 衰减s + 空间定位), A3 Voice (基频Hz + 气声比% + 语速字/秒 + 混响 RT60) — banning emotion adjectives in favor of physical parameters (配乐 / 音效 / 配音) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【分镜模式·规避审查 / Storyboard · Dodge】触发词:规避审查 / 过审 / 绕过敏感词 / 暴力裸露改写 / 轮廓化 / 力学矢量改写 / content filter / dodge / safe rewrite。Use when output text risks tripping content filters with explicit anatomical terms, action verbs (thrust/penetrate), graphic injury, or nudity — rewrite via contour/silhouette, force vectors (impact 200N, compression 3cm), pressure / velocity / amplitude, fabric silhouette, shallow DoF blur, or steam diffusion (规避审查) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【分镜模式·风格 / Storyboard · Style】触发词:风格 / 配色 / 色板 / 色彩 / 光源 / 阴影 / 色调 / 美术风格 / palette / hex / lighting / shadow。Use when a scene needs a hard palette decomposition — dominant [hex] + accent [hex] in ratio ≥ 7:3, light source typed as rim / fill / key with angle and intensity %, and shadow depth expressed as percentage of frame in shadow (配色 / 光源 / 阴影) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【分镜模式·视觉 / Storyboard · Visual】触发词:视觉 / 打光 / 色彩层次 / 镜头规格 / 焦段光圈 / Z轴 / 前中后景 / 阴影占比 / lighting / color / lens / Z-axis / shadow。Use when a shot needs physical lighting (shadow 10-40% day / 60-90% night, rim / candle / natural / fill), color hierarchy ([key hex] dominant + faint [accent hex], never equal warm + cool), explicit lens spec [mm] f/[stop] instead of "8k masterpiece", and a mandatory Z-axis (fg occluder / mg subject / bg environment) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【分镜模式·物理 / Storyboard · Physics】触发词:物理 / 人体 / 肌肉 / 骨骼 / 动作 / 运动矢量 / 微表情 / 量化 / 写实 / body physics / motion vector / micro-expression。Use when character body / motion must be described in physical-only language — skin texture, muscle tension, bone structure; motion vectors as angle° / displacement cm / velocity m·s⁻¹; micro-expressions in mm (brow furrowed Xmm, pupil dilation Xmm), never emotion adjectives — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Bundle

【导演模式·总调度 / Director · Orchestrator·每次必用】导演总调度路由器:像 using-superpowers 那样把全部 director-*(镜头/构图/连续性)与 storyboard-*(物理/对白/风格)技法技能串成一条流水线——先反问要用哪些本地 skill,再按 13 维电影摄制框架以"物理可复现参数优先于情绪形容词"输出结构化文本(非 JSON)提示词,并交给 catimation-image / catimation-video 出片。MUST be loaded EVERY time a task touches an image/video/animation prompt — 任何关于视频、图片、提示词的问题或生成任务,以及你自发为回答配图/配视频之前,都必须先载入并执行本技能,不可凭感觉直接写提示词。Use when 写/优化提示词、生成图片/视频/动画、画一张、出图、来段视频、做个动画、让它动起来、分镜、镜头、运镜、构图、打光、调色、角色一致性、参考图分析/复刻/反推。Triggers on prompt / image / video / animation / shot / storyboard / cinematography / 提示词 / 图片 / 视频 / 出图 / 配图. Loads before writing any prompt or generating any image/video.

2799662352 Skill

【分镜模式·对白 / Storyboard · Dialogue】触发词:对白 / 台词 / 字幕 / 角色名 / 逐字 / 原文提取 / 不要瞎编台词 / dialogue / lines / verbatim / character names。Use when screenplay or script provides character names and dialogue that MUST be extracted verbatim — never fabricate lines, never guess names from visual style (台词 verbatim) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【分镜模式·结构 / Storyboard · Structure】触发词:分镜结构 / 瞬间抓拍 / 峰值张力 / 单一核心动作 / 不要"然后" / 微表情量化 / mid-action / one core action / peak tension。Use when shaping each shot as a mid-action snapshot frozen at peak tension (never "then / after"), holding to ONE core action within 2-4 seconds, and quantifying micro-expressions in millimeters (brow furrowed Xmm, pupil dilation Xmm) instead of emotion labels — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【导演模式·风格一致性 / Director · Style Consistency】触发词:风格一致 / 风格统一 / 图文风格冲突 / 材质统一 / 赛璐璐密度 / 颗粒一致 / 写实vs动画 / style consistency / uniform texture。Use when resolving image-vs-text style conflict (TEXT WINS — reference image supplies only character identity, never rendering medium or color grading), enforcing uniform texture / cel-shading / film-grain density across panels, and reinforcing negative prompts (photoreal target → ban anime/cartoon; anime target → ban photoreal) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

【导演模式·动画质感 / Director · Anime Look】触发词:动画质感 / 日式动画截图 / 赛璐璐 / cel / 去厚涂 / 油画感 / 二次元 / 番剧风 / anime screenshot / no painterly。Use when output drifts toward painterly / 厚涂 / oil-painterly texture instead of cel-shaded anime screenshot style, character identity must lock across panels (hair / outfit / accessories), or injecting JSON instruction blocks with Color Harmony 1.5 and No-Painterly 1.8 weights (日式动画截图) — applies to image-generation models (Midjourney, DALL-E, FLUX, Stable Diffusion, Imagen, Ideogram, Recraft), video-generation models (Sora, Veo, Runway, Kling, Seedance, Hailuo, Higgsfield, Hunyuan), screenplays, scripts, storyboards, AI video, AI image, 提示词, 视频模型, 图像模型, 写剧本, 脚本, 分镜.

2799662352 Skill

用于通过负向提示词和边界限制反向控制 AI 画质,解决默认网红脸、塑料锐化、脏纹理、梦幻=烟雾、高级=金粉、自动补齐背景等误解机制。适用于 negative prompt、--no、禁止词、反向控制、抽象词拆解、视觉公式、缺省词、自动补齐、画质真实感、Midjourney/Stable Diffusion/NanoBananaPro。