English · العربية · Español · Français · 日本語 · 한국어 · Tiếng Việt · 中文 (简体) · 中文（繁體） · Deutsch · Русский
English · 中文 (简体) · 中文（繁體） · 日本語 · 한국어 · Français · Español · Deutsch · Italiano · Português · Русский · 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
LALACHAN daily story/video operators can start from
docs/lalachan-story-video-generation-handoff.md.
Fast Start
git clone [email protected]: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 |
|---|
overleaf-local-sync | Reconciles local and Overleaf manuscript edits, preserves collaborator changes, and verifies incremental text and figure updates. | Three-way comparison, source ZIPs, Git, browser/CDP, PDF verification |
novnc-webapp-control | Operates and validates first-party webapps through a dedicated, observable noVNC browser while preserving visible user controls and paid-action guards. | Xvfb, x11vnc, noVNC, Chrome/CDP, Playwright, screenshots, artifact probes |
store-login-session | Reuses Apple/Google store logins and recovers remembered-account sign-in before interrupting a release. | Existing noVNC/CDP profile, observed Continue controls, private credential handoff, submission reconciliation |
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 |
musia-lalachan-mv-workflow | Creates song-first LALACHAN/Xiaoyunque MV handoffs from reviewed Musia tracks, including full-song story MVs, chorus/highlight cuts, and song-locked audio replacement. | Musia audio, Xiaoyunque prompts, timestamped MV segments, ffmpeg muxing, LazyEdit-ready outputs |
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. |
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
Local paths should live in an ignored local config file:
cp .config/lazyskills.env.example .config/lazyskills.local.env
$EDITOR .config/lazyskills.local.env
set -a
. .config/lazyskills.local.env
set +a
Preferred no-copy external pack use:
export AGINTIFLOW_SKILL_PACKS="$LAZYSKILLS_ROOT"
aginti skills "npm publishing"
aginti skills "PocketPolyglot"
Project-local copy when a single project needs pinned/customized skills:
cd /path/to/project
python3 $LAZYSKILLS_ROOT/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 $AGINTIFLOW_ROOT
node --input-type=module - <<'NODE'
import fs from "node:fs/promises";
import {
buildSkillPackFromMarkdown,
enableSkillMeshSkill,
installSkillPack,
} from "./src/skillmesh.js";
const skillsRoot = process.env.LAZYSKILLS_ROOT;
if (!skillsRoot) throw new Error("Set LAZYSKILLS_ROOT before running this example.");
const content = await fs.readFile(`${skillsRoot}/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/codex-session-recovery ~/.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/jlceda-mcp-automation ~/.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 selected skills/<id>/SKILL.md.
- Gemini: start from
GEMINI.md, then skills.json, then selected SKILL.md.
- GitHub Copilot: repository instructions live in
.github/copilot-instructions.md and .github/instructions/lazyskills.instructions.md.
- Generic agents: use
skills.json as 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/codex-session-recovery/SKILL.md
python3 -m py_compile skills/codex-session-recovery/scripts/*.py
python3 skills/codex-session-recovery/scripts/test_codex_session_recovery.py
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/jlceda-mcp-automation/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
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
