liteharness-plugin

Agents Overflow — ask, answer, browse, search questions. Triggers on 'ao', 'agents overflow', or any AO-related request (posting questions, answering, browsing, searching).

Generate or update architecture documentation for any project. Dispatches scout agents to explore the codebase, then writes structured architecture docs with an INDEX. Triggers on 'generate arch docs', 'create architecture docs', 'update arch docs', 'document this project', 'arch-gen', 'generate documentation'.

Canonical architecture reference for LiteSuite mega-app and ecosystem. Quick-lookup index into 15 architecture docs covering panels, services, MCP tools, voice, harness, arena, and more. Triggers on 'arch', 'architecture', 'show architecture', 'arch docs'.

Use when creating posters, visual art, graphic designs, or static visual compositions as .png or .pdf. Triggers on 'create a poster', 'design a poster', 'make artwork', 'create visual art', 'design something beautiful', 'graphic design'.

Reverse-engineer any codebase, plugin, or repo into an evidence-first case study with file:line citations for every claim. Pattern extraction for future reference — not a bug hunt.

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

Translate ComfyUI workflows into LiteImage API calls. Use when the user provides a ComfyUI workflow JSON, mentions ComfyUI, asks about node graphs, wants to convert or replicate a ComfyUI pipeline, or references ComfyUI templates. Also use when the user asks how to do something 'like in ComfyUI' or wants to run a workflow locally. Triggers on 'comfy', 'ComfyUI', 'comfyui', 'workflow JSON', 'node graph', 'convert this workflow', 'translate this workflow', 'run this workflow', 'I have a workflow', 'KSampler', 'LoraLoader', 'exported workflow', 'comfy template', 'what this comfy workflow does', 'equivalent in LiteImage', 'how do I do that here', 'make it work here'.

Turn any MCP server or SKILL.md file into a standalone CLI using LiteCLI. Use when the user wants to compile an MCP server, skill file, or directory of skills into runnable CLI commands.

Spawn expert polymathic thinking agents for deep analysis. Use proactively when the user wants expert opinions, multi-perspective review, or mentions specific thinkers by name. Triggers on 'consult polymaths', 'ask the polymaths', 'polymath review', 'expert opinions on', 'what would the polymaths say', 'get Feynman to review', 'ask Carmack', 'what would Jobs think', 'multi-perspective review', 'diverse expert perspectives', 'run it by the polymaths', 'task the polymaths', 'debate this architecture', 'expert analysis'. Also trigger when the user names ANY polymathic agent (Feynman, Carmack, Jobs, Munger, Shannon, Linus, Tesla, etc.). Internal thinking agents — different from /consult which queries external LLM APIs.

Query external LLM APIs (GPT, Gemini, LM Studio) and Claude CLI for opinions. Only invoke when the user explicitly says 'consult', 'ask GPT', 'ask Gemini', or specifically requests external model comparison. For expert analysis, code review, or architecture opinions, prefer /consult-polymaths instead — it uses internal agents and works without any external services.

Find, search, and summarize Claude Code conversations. BM25 + semantic (vector) + hybrid search across indexed messages. Uses all-MiniLM-L6-v2 embeddings (sentence-transformers). USE PROACTIVELY when you need historical context — don't rg through JSONL files manually. Triggers on 'find conversation', 'search conversations', 'search convos', 'convo search', 'what conversation did we', 'which session did we', 'when did we build', 'find where we discussed', 'remember when we', or when a bare 8-char hex ID is given. Also use when the user references past work and you need to locate it. Use '--mode semantic' for conceptual queries, '--mode hybrid' for best results.

LiteSuite runtime debugging — set up disk logging, Monitor tail, rebuild loop. Use when debugging Electron webview, preload, bridge, IPC, or any runtime issue where reading code alone won't reveal the problem. Triggers on 'debug litesuite', 'ls-debug', 'debug the panel', 'debug codex', 'debug claude panel', 'why is it not loading', 'webview blank', 'bridge not connecting'.

花叔Design（Huashu-Design）——用HTML做高保真原型、交互Demo、幻灯片、动画、设计变体探索+设计方向顾问+专家评审的一体化设计能力。HTML是工具不是媒介，根据任务embody不同专家（UX设计师/动画师/幻灯片设计师/原型师），避免web design tropes。触发词：做原型、设计Demo、交互原型、HTML演示、动画Demo、设计变体、hi-fi设计、UI mockup、prototype、设计探索、做个HTML页面、做个可视化、app原型、iOS原型、移动应用mockup、导出MP4、导出GIF、60fps视频、设计风格、设计方向、设计哲学、配色方案、视觉风格、推荐风格、选个风格、做个好看的、评审、好不好看、review this design。**主干能力**：Junior Designer工作流（先给假设+reasoning+placeholder再迭代）、反AI slop清单、React+Babel最佳实践、Tweaks变体切换、Speaker Notes演示、Starter Components（幻灯片外壳/变体画布/动画引擎/设备边框）、App原型专属守则（默认从Wikimedia/Met/Unsplash取真图、每台iPhone包AppPhone状态管理器可交互、交付前跑Playwright点击测试）、Playwright验证、HTML动画→MP4/GIF视频导出（25fps基础 + 60fps插帧 + palette优化GIF + 6首场景化BGM + 自动fade）。**需求模糊时的Fallback**：设计方向顾问模式——从5流派×20种设计哲学（Pentagram信息建筑/Field.io运动诗学/Kenya Hara东方极简/Sagmeister实验先锋等）推荐3个差异化方向，展示24个预制showcase（8场景×3风格），并行生成3个视觉Demo让用户选。**交付后可选**：专家级5维度评审（哲学一致性/视觉层级/细节执行/功能性/创新性各打10分+修复清单）。

Spawn a local Devstral coding agent running on LM Studio. Use when asked to '/devstral', 'use devstral', 'ask devstral', 'local agent', 'run this on devstral', 'devstral this'. Routes tasks through the local model_router.py proxy at port 5111 which transparently forwards devstral requests to LM Studio while keeping Anthropic calls working normally.

Memory consolidation — reflective pass over memory files to synthesize, prune, and organize durable memories. Triggers on 'dream', 'consolidate memory', 'clean up memory', 'organize memories', 'memory maintenance'.

Use when the user launches Claude from inside the LiteEVA panel, wants to create or work on a video project, or asks about agentic video editing. Subcommands — remotion, voiceover, music, ffmpeg. Triggers on 'LiteEVA', 'open EVA', 'launch EVA', 'video project', 'start a video', 'new video', 'EVA panel', 'video editor', 'create video', 'work on video', 'remotion', 'composition', 'render video', 'voiceover', 'TTS', 'generate narration', 'generate music', 'soundtrack', 'ACE-Step', 'ffmpeg', 'trim video', 'compress video', 'platform export'.

Use when discovering, searching for, or installing new agent skills from the open ecosystem. Triggers on 'find a skill', 'is there a skill for', 'search for skills', 'install a skill', 'search skills.sh', 'find-skill-sh'. Covers both npx skills CLI and the skills.sh ecosystem.

Use when generating AI images or videos — marketing visuals, hero banners, social media assets, product mockups, ad creative, animated backgrounds, exploding views, 3D renders, scroll-synced video assets. Triggers on 'generate an image', 'create an image', 'gen-image', 'make me an image', 'image gen', 'marketing image', 'product mockup', 'hero image', 'generate a video', 'create a video', 'gen-video', 'make me a video', 'video gen', 'animated background', 'exploding view', 'veo', 'video asset', 'scroll animation video', 'chatgpt image', 'gpt image', 'chatgpt flag'.

LiteSuite Generative UI — render interactive widgets inline in Frontier Chat / Sentinel Chat. Use prompt_widget when you need the user's input to continue (confirmations, forms, picks). Use render_widget for fire-and-forget displays (charts, stat cards, dashboards). Triggers on 'render widget', 'prompt widget', 'ls-generative-ui', 'show me a chart', 'ask the user', 'confirm before', 'show a form', or whenever a visual or interactive answer is clearer than prose.

Onboarding wizard — interviews the user, discovers their cognitive architecture, and generates a fully personalized 5-tier orchestration environment. One command, the harness thinks like you.

Deep usage insights across ALL Claude Code sessions (full history, not just recent). Scans every JSONL conversation file, merges with facets and session-meta, then generates a beautiful visual HTML report using the mockup skill. Use when you want comprehensive analysis covering months of sessions, not just the last 50-60.

Use when searching for a reference LLM app, AI project, or code example to learn from before building something new. Triggers on 'find an app like', 'show me an example of', 'reference project for', 'k-find-app'. Searches a local catalog of 207 apps then analyzes real GitHub source code.

Team Lead bootstrap for LiteSuite — spawns and manages workers, reports to orchestrator.

Eratosthenes — The Librarian. Dispatches 4 Haiku scout sub-agents to verify architecture docs, skill catalogs, memory files, and dead references against actual code state. Synthesizes reports and auto-fixes all drift. Applies the cognitive architecture of the chief librarian of Alexandria.

Use when needing to discover, search, list, or describe available skills, commands, and agents across all codebases and locations. Triggers on 'library', 'list skills', 'what skills do I have', 'find a skill', 'show all agents', 'what commands are available', 'list my agentics', 'catalog', 'describe this agent', 'what does this skill do'. Also trigger on 'do I have a skill for [topic]', 'is there an agent for [topic]', 'what commands exist in [project]', 'describe [agent-name]', 'tell me about [skill-name]', 'what polymathic agents exist', 'list commands in <project>', or any question checking whether a specific skill, command, or agent exists or requesting an inventory of available agentics.

LiteHarness agent orchestration — spawn, name, message, and control Claude Code sessions. Headless (ConPTY) and headed (UIAutomation) modes. Use when spawning agents, checking inbox, sending messages, discovering agents, controlling terminals, or when hooks fail. Triggers on 'liteharness', 'spawn', 'check inbox', 'send a message to', 'discover agents', 'who is online', 'watch inbox', 'start liteharness', 'send-input', 'read-output'.

Generate rich Textual TUI + Click CLI applications from any software. Wraps arbitrary software in a professional terminal interface with tabbed layout, sidebar, themes, and shared backend. Use when asked to create a TUI, terminal interface, or CLI wrapper for software.

Use when the user wants screen awareness or visual context about what they're doing. Triggers on 'what's on my screen', 'start litewatch', 'watch my screen', 'screen awareness'. Launches the LiteWatch TUI to capture screenshots via local VLM and periodically checks screen_context.txt.

Use when you need to process, summarize, or extract information from large content without bloating context. Triggers on 'summarize this', 'describe this screenshot', 'compress context', 'offload to local model', 'local-lens'. Routes through local LM Studio models. For planning use /plan-w-quizmaster, for expert analysis use /consult-polymaths.

Use when decomposing a task into parallel subtasks with structured subagent spawning. Triggers on 'max parallel', 'parallelize this', 'spawn subagents', 'decompose and parallelize', 'break this into parallel tasks', 'recursive decomposition'. NOT for simple agent swarms (use /max-swarm).

Use when spawning multiple coding agents to work on independent files or modules simultaneously. Triggers on 'max-swarm', 'swarm this', 'swarm it', 'spawn a swarm', 'coding swarm', 'one agent per file', 'throw agents at this'. NOT for structured decomposition (use /max-parallel).

Use when creating visual mockups, screenshot layouts, page previews, prototype arrangements, data visualizations, or any visual design work. Outputs to a Next.js mockup project (user-supplied). Triggers on 'mockup', 'create a mockup', 'layout mockup', 'screenshot preview', 'page preview', 'arrange screenshots', 'visual prototype', 'visualize this', 'make this pretty', 'visualize this markdown', 'visualize this document'.

Use this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.

Use when the user wants to plan something — load immediately when planning intent is expressed, don't ask first. Triggers on 'I want to plan', 'let's plan', 'plan something out', 'help me plan', 'plan it out', 'quiz me on the requirements', 'plan-w-quizmaster'. NOT a mandatory gate on implementation. For expert opinions use /consult-polymaths.

Triggers on 'autonomous plan', 'auto-plan', 'sonnet plan', or 'plan-w-quizmaster-sonnet' — also when user says they want a plan made while they're away or without their involvement. Opus runs the Quizmaster interrogation while a spawned Sonnet agent answers as the developer, grounded in codebase context. The user is an observer only — no questions asked of them. Use /plan-w-quizmaster (interactive) when the user wants to answer questions themselves.

Use when creating E2E Playwright tests or browser automation tests with visual verification. Triggers on 'e2e test', 'playwright test', 'end to end test', 'browser test with screenshots', 'visual verification test'.

Rebuild all Lite Suite apps with pnpm dist, regenerate SHA-256 hashes from portable zips, update litesuite.dev manifest.json, and push GitHub releases. Use after icon changes, version bumps, or any change that requires rebuilding the installers.

Lite Suite Release Pipeline. Use when releasing LiteSuite or LiteImage, updating litesuite.dev manifests, building release installers, creating GitHub releases, or performing gated Lite Suite release checks.

Reviewer bootstrap for LiteSuite — code review and quality assessment.

PROACTIVE BEHAVIOR — not primarily user-triggered. Opus should automatically dispatch Haiku sub-agents for simple research: reading files, listing directories, web searches, doc summaries, factual lookups. Saves Opus tokens by offloading grunt work to the cheapest model with its own separate usage pool. Use Haiku for facts, Sonnet polymaths for reasoning, Opus for decisions. Triggers on 'scout', 'send a scout'. Proactively used without user asking.