LazySkills
lachlanchen/LazySkills
Skills portáveis e reutilizáveis para agentes executarem trabalho real
Ver no GitHubBiblioteca portável de skills reutilizáveis para agentes de IA (Claude, Gemini, Copilot, AgInTiFlow) que precisam executar workflows complexos. Compacta procedimentos comprovados como playbooks SKILL.md com scripts, referências e métodos de validação determinísticos para tarefas de automação, publicação e browser automation.
Instalação
npx convez-ai lachlanchen/LazySkills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automação de workflows de publicação de conteúdo e vídeos
- Browser automation com validação e logging estruturado
- Integração de agentes IA com ferramentas externas via skills modulares
Veredito

LazySkills é uma abordagem pragmática para compartilhar procedimentos operacionais entre agentes diferentes, separando lógica de agent-specific prompting. Ideal para quem constrói sistemas que precisam de memory externa e procedimentos determinísticos, mas a curva de aprendizado em configuração cross-platform é moderada.
README
A tradução para o português deste README ainda está na fila.
English · ç®€ä½“ä¸æ–‡ · ç¹é«”䏿–‡ · 日本語 · 한êµì–´ · Français · Español · Deutsch · Italiano · Português · РуÑÑкий · العربية
LazySkills
Portable skills for agents that need to act, verify, and finish real work.
LazySkills is a reusable skill library by lachlanchen for the LazyingArt ecosystem. It packages proven workflows as compact SKILL.md playbooks with optional scripts, references, and validation methods.
Website: lachlanchen.github.io/LazySkills
Fast Start
git clone git@github.com:lachlanchen/LazySkills.git
cd LazySkills
python3 scripts/lazyskills.py validate
python3 scripts/lazyskills.py list
Use without copying in AgInTiFlow:
export AGINTIFLOW_SKILL_PACKS="$PWD"
aginti skills "npm publishing"
Install locally for Codex or Claude:
python3 scripts/lazyskills.py install --platform codex
python3 scripts/lazyskills.py install --platform claude
Why LazySkills
Modern agents are strongest when they have reliable working memory outside the chat window: task-specific procedures, deterministic scripts, and evidence-based completion checks. LazySkills keeps those pieces organized so different agents can reuse the same operational knowledge instead of rediscovering it every session.
The repository is intentionally platform-neutral. A skill should be useful to AgInTi, Codex, Gemini, GitHub Copilot, Claude, or any local-tool agent harness that can read files and run commands.
Supported Agents
| Platform | How to use LazySkills |
|---|---|
| AgInTi / AgInTiFlow | Prefer AGINTIFLOW_SKILL_PACKS=/path/to/LazySkills; copy to project .aginti/skills/ only for project-local overrides. |
| Codex | Install folders into ~/.codex/skills/ with scripts/lazyskills.py install --platform codex. |
| Claude agents | Use CLAUDE.md as the repo entrypoint or copy folders into a Claude-readable skill directory. |
| Gemini agents | Use GEMINI.md as the repo entrypoint and skills.json for skill discovery. |
| GitHub Copilot agents | Use .github/copilot-instructions.md and .github/instructions/lazyskills.instructions.md. |
| Generic local-tool agents | Read skills.json, then the selected skills/<id>/SKILL.md, and run bundled scripts under normal policy. |
Available Skills
| Skill | What it does | Key tools |
|---|---|---|
lalachan-xyq-browser-video | Generates and monitors Xiaoyunque videos through the logged-in browser UI, with visible validation before paid submission. | Chrome/CDP, upload verification, prompt fill, thread watcher, MP4 download fallback |
lazyedit-publish-workflow | Publishes LazyEdit videos and AI-generated LALACHAN/RARACHAN videos through AutoPubMonitor and AutoPublish, including subtitle correction and queue monitoring. | LazyEdit CLI/API, AutoPubMonitor, lazyingart SSH, tmux, Shipinhao, YouTube, Instagram |
media-transcription-report | Transcribes audio or video into timestamped Markdown with optional WhisperX speaker diarization, then turns transcripts or source notes into neutral third-person Markdown/TeX/PDF reports. | Whisper, WhisperX, ffmpeg, speaker diarization, Markdown, XeLaTeX/PDF compilation |
npm-publishing | Packages, publishes, and verifies npm packages while handling 2FA, token files, trusted publishing, and install smoke tests safely. | npm, GitHub Actions OIDC, temp .npmrc, registry verification |
publish-repo | Publishes local git repositories to GitHub with profile-style multilingual READMEs, LazyingArt banner/support panels, sponsors, citation metadata, safe commits, remote creation, homepage, description, and topics. | git, GitHub CLI, repo metadata, i18n README, funding config, CITATION.cff, Codex/AgInTi handoff |
ocr-book-polisher | Converts scanned or image-heavy books into corrected, publishable Markdown/TeX/PDF while preserving figures, captions, structure, and evidence checks. | OCR, page-aware Markdown, strict TOC validation, TeX/PDF compilation |
libgen-safe-book-browser | Opens LibGen search/detail pages through Chrome/CDP while blocking Trip.com and ad redirects, with exact metadata and PDF > EPUB > other candidate selection. | LibGen API, Chrome/CDP Fetch interception, redirect guard |
pocketpolyglot-bookmaker | Builds LinguaLeaf/PocketPolyglot-style multilingual interlinear pocket books with durable JSON, ruby/pinyin/furigana, grammar roles, and bidirectional PDFs. | EPUB/PDF extraction, chunk manifests, JSON validation, tmux workers, XeLaTeX |
transcript-video-section-splitter | Transcribes videos, derives topic-based section boundaries, and splits edited or source videos into named clips with manifests. | Whisper/WhisperX, content sections, ffmpeg, clip manifest |
video-face-image-replacement | Covers or replaces one or more detected video faces with supplied or generated image assets, including multi-person animal masks with stable identity tracking. | InsightFace/SCRFD, identity maps, RetinaFace/MediaPipe options, alpha compositing, ffmpeg |
aginti-agentlink | Coordinates multiple agent sessions across machines, repos, tools, hardware, and APIs using safe handoffs and action contracts. | peer maps, handoff packets, private session mirrors, status probes, evidence bundles |
kv260-metavision-lab | Operates and maintains the AMD Kria KV260 + Prophesee Metavision lab, including desktop launchers, custom viewer, recording API, native viewer recovery, Windows X11 control center, and file transfer. | PetaLinux, Prophesee, Metavision, V4L2, X11/Matchbox, SSH X11, event recording |
kv260-windows-arduino | Coordinates KV260 event recording with the Windows host and USB Arduino light controller, including LAN identity, COM-port checks, future Arduino APIs, and cross-session handoffs. | KV260 API, Windows SSH, Arduino CLI, COM ports, DualLampHI, private session mirrors |
kicad-mcp-pcb-design | Installs and uses KiCad with MCP/agent automation to inspect old PCB projects, generate boards, save datasets, validate DRC, export Gerbers/STEP, and render previews. | KiCad 10, kicad-cli, pcbnew, MCP stdio, Gerber/STEP/render export |
dual-led-constant-power | Calibrates and controls dual Arduino-driven LED branches with MOS modules and INA219 monitoring for smooth constant-power crossfades. | Arduino UNO, YYNMOS MOS, INA219, inverse LUT, CSV/PNG telemetry |
parametric-cad-design | Designs, revises, validates, documents, and renders mechanical CAD parts with versioned artifacts and measured print-fit compensation. | OpenSCAD, CadQuery/build123d/OCP, FreeCAD, Blender, STEP/STL/DXF/SVG/PDF, mesh checks |
Skill Anatomy
skills/
skill-name/
SKILL.md # trigger conditions and core workflow
scripts/ # deterministic helpers for fragile steps
references/ # deeper notes loaded only when needed
skills.json # machine-readable skill inventory
AGENTS.md # general/Codex-style repo instructions
CLAUDE.md # Claude project instructions
GEMINI.md # Gemini project instructions
.github/ # Copilot instructions and sponsorship metadata
docs/ # general workflow documentation
i18n/ # localized README summaries
This structure keeps the skill itself concise while still preserving the tools and details needed for hard tasks.
Install
AgInTiFlow
Preferred no-copy external pack use:
export AGINTIFLOW_SKILL_PACKS="/home/lachlan/ProjectsLFS/LazySkills"
aginti skills "npm publishing"
aginti skills "PocketPolyglot"
Project-local copy when a single project needs pinned/customized skills:
cd /path/to/project
python3 /home/lachlan/ProjectsLFS/LazySkills/scripts/lazyskills.py install --platform aginti --scope project npm-publishing
aginti skills "npm publishing"
SkillMesh import still works when you want reviewed/enabled SkillMesh copies. Example for npm-publishing:
cd /home/lachlan/ProjectsLFS/Agent/AgInTiFlow
node --input-type=module - <<'NODE'
import fs from "node:fs/promises";
import {
buildSkillPackFromMarkdown,
enableSkillMeshSkill,
installSkillPack,
} from "./src/skillmesh.js";
const content = await fs.readFile("/home/lachlan/ProjectsLFS/LazySkills/skills/npm-publishing/SKILL.md", "utf8");
const pack = await buildSkillPackFromMarkdown(content, { valueScore: 92 });
await installSkillPack(pack, { enabled: true });
await enableSkillMeshSkill("npm-publishing", true);
console.log(JSON.stringify({ ok: true, installedSkills: pack.skills.map((skill) => skill.id), packHash: pack.packHash }, null, 2));
NODE
Codex
python3 scripts/lazyskills.py install --platform codex
Manual equivalent:
cp -R skills/lalachan-xyq-browser-video ~/.codex/skills/
cp -R skills/lazyedit-publish-workflow ~/.codex/skills/
cp -R skills/media-transcription-report ~/.codex/skills/
cp -R skills/npm-publishing ~/.codex/skills/
cp -R skills/publish-repo ~/.codex/skills/
cp -R skills/ocr-book-polisher ~/.codex/skills/
cp -R skills/pocketpolyglot-bookmaker ~/.codex/skills/
cp -R skills/transcript-video-section-splitter ~/.codex/skills/
cp -R skills/video-face-image-replacement ~/.codex/skills/
cp -R skills/aginti-agentlink ~/.codex/skills/
cp -R skills/kv260-metavision-lab ~/.codex/skills/
cp -R skills/kv260-windows-arduino ~/.codex/skills/
cp -R skills/kicad-mcp-pcb-design ~/.codex/skills/
cp -R skills/dual-led-constant-power ~/.codex/skills/
cp -R skills/parametric-cad-design ~/.codex/skills/
Claude, Gemini, Copilot, Generic Agents
- Claude: start from
CLAUDE.md, then selectedskills/<id>/SKILL.md. - Gemini: start from
GEMINI.md, thenskills.json, then selectedSKILL.md. - GitHub Copilot: repository instructions live in
.github/copilot-instructions.mdand.github/instructions/lazyskills.instructions.md. - Generic agents: use
skills.jsonas the index and keep skill folder structure intact.
For folder-copy agents:
python3 scripts/lazyskills.py install --platform claude
python3 scripts/lazyskills.py install --platform generic --dest ./agent-skills npm-publishing
Quick Validation
Check browser automation helpers and skill file layout before using them:
python3 scripts/lazyskills.py validate
python3 scripts/lazyskills.py list --json
python3 skills/lalachan-xyq-browser-video/scripts/xyq_cdp_browser.py --help
python3 skills/lalachan-xyq-browser-video/scripts/xyq_chrome/watch_thread_dom_download.py --help
sed -n '1,40p' skills/lazyedit-publish-workflow/SKILL.md
test -f skills/media-transcription-report/SKILL.md
test -f skills/npm-publishing/SKILL.md
test -f skills/publish-repo/SKILL.md
bash -n skills/publish-repo/scripts/publish_repo.sh
test -f skills/ocr-book-polisher/SKILL.md
test -f skills/pocketpolyglot-bookmaker/SKILL.md
test -f skills/transcript-video-section-splitter/SKILL.md
test -f skills/video-face-image-replacement/SKILL.md
test -f skills/aginti-agentlink/SKILL.md
test -f skills/kv260-metavision-lab/SKILL.md
test -f skills/kv260-windows-arduino/SKILL.md
test -f skills/kicad-mcp-pcb-design/SKILL.md
test -f skills/dual-led-constant-power/SKILL.md
test -f skills/parametric-cad-design/SKILL.md
bash -n skills/kv260-metavision-lab/scripts/kv260_metavision_probe.sh
bash -n skills/kv260-windows-arduino/scripts/fetch-windows-codex-session.sh
bash -n skills/kv260-windows-arduino/scripts/kv260-lab-status.sh
bash -n skills/kv260-windows-arduino/scripts/kv260-record-once.sh
bash -n skills/kv260-windows-arduino/scripts/windows-arduino-probe.sh
aginti skills "npm publishing"
Documentation
- LazySkills website source
- Platform support
- Xiaoyunque browser video publishing
- Smooth Xiaoyunque video generation runbook
- Xiaoyunque continue confirmation and protected download runbook
- LazyEdit publish runbook
- Publish repo skill handoff
- Reusable language header
Detailed, project-specific run logs should live in the project that produced them. LazySkills keeps only reusable procedures and general examples.
Design Principles
- Keep skills small enough to load quickly.
- Put fragile operations into scripts instead of rewriting them repeatedly.
- Validate before claiming completion.
- Keep API keys, cookies, browser profiles, generated videos, and private run artifacts out of the repo.
- Prefer general methods over hard-coded one-off task state.
LazyingArt
Built by Lachlan Chen for creative AI workflows across storytelling, multilingual publishing, browser agents, and practical automation.
Main site: lazying.art
Creative goods: buy.lazying.art
Support
| Donate | PayPal | Stripe |
|---|---|---|