One command drops a tested, versioned skill into your agent. No more ad-hoc prompts. The same rigorous steps run every time.
Skills are independent files. Works with Claude Code, Cursor, Windsurf, GitHub Copilot, OpenAI Codex, OpenCode, Google Antigravity (install.sh:23).
This catalog release adds three skills. user-test tests an app, API or CLI end to end through its real interface, records evidence for every finding, and fixes simple and medium findings behind checkpoints that a fresh verifier confirms, leaving the fixes uncommitted. product-launch scores 50 launch sites for your product, checks each shortlisted site's live submit, pricing and rules pages, and after you approve the plan writes paste-ready posts, pitches and a dated checklist; it never signs up, submits, votes, pays or posts for you. rex-agent brings herdr-agent's fleet workflow to the Rex terminal. Every skill also gains an optional, propose-only ## Self-Improve check that suggests one edit to itself when a run shows a lasting mismatch with how you work, and changes nothing until you say yes. devops-pipeline now sets up and audits fast CI at no extra cost, appstore-assets renders the looping App Store header and search videos and landscape sets, landing-page-generator ends with a built static site and its Privacy, Terms, Changelog and Docs pages, and appstore-review-checker follows Apple's App Review Guidelines of October 16, 2026. lstack 0.7.0 adds the three new skills for 46. Individual skills keep their own versions; the catalog release does not reset them to 4.3.0.
Upgrading from 4.2.0:
- Add the new skills:
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill user-test --skill product-launch --skill rex-agent. Skills you already installed are unchanged; reinstall them to get the## Self-Improvesection and the other updates. - If you installed lstack from this repository's marketplace, run
claude plugin marketplace update lstack, thenclaude plugin update lstack@lstack --scope user, and restart Claude Code to get 0.7.0. ZIP users downloadlstack-0.7.0.zip,lstack-claude-0.7.0.ziporlstack-codex-0.7.0.zipfrom the v4.3.0 release and install them as before. - Contributors: the
skills/plugin folder now holds 512 files, the plugin directory's limit, so a change that adds a file underskills/must merge or remove another one (eval fixtures still go ineval-fixtures/<skill>/; see CONTRIBUTING.md).
Earlier upgrade guidance (from 4.1.0):
- Add github-pages-publish:
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill github-pages-publish. - Contributors: put eval input fixtures in
eval-fixtures/<skill>/, never underskills/, and pointevals/evals.jsonfilesentries at../../eval-fixtures/<skill>/(see CONTRIBUTING.md).
Earlier upgrade guidance (from 4.0.0):
- If you load lstack from a clone with
--plugin-dir, point it at the clone'sskills/folder instead of the repository root:claude --plugin-dir /absolute/path/to/checkout/skills. ZIP downloads install as before. - If you installed lstack from this repository's marketplace, run
claude plugin marketplace update lstack, thenclaude plugin update lstack@lstack --scope user, and restart Claude Code.
Earlier upgrade guidance (from 3.0.0):
- Install ship:
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill ship, then replace/release-managerinvocations with/ship. - Review and remove the obsolete installed
release-managercopy yourself; reinstalling does not remove it automatically, and both names can compete for triggers. /ship [X.Y.Z|major|minor|patch]runs autonomously by default. Use/ship --no-autoto keep step-by-step confirmations.
Earlier upgrade guidance (from 2.0.0):
herdr-agent-commsis now herdr-agent. Reinstall under the new name and update prompts; installers do not automatically remove the old installed copy, which can still trigger.plan-to-issuesmoved to luongnv89/idd; see the installation note. Its epic output is now a static plan map with status maintained through native GitHub sub-issues. Update parsers expecting- [x] #N — <id>to accept- #N — <id>.- cleanup-project replaces
branch-inspector.convert-documents-to-markdown,fork-upstream-sync,opencode-sandbox,opencode-handoff,subagent-creator, and thewebsite-clonersuite are retired from this catalog. Existing installed copies are not automatically removed.
See CHANGELOG.md for the full release history.
Validate this runbook:
./scripts/validate-install.sh --check
lstack 0.7.0 packages the canonical skills/ tree as one
Claude Code plugin: 44 top-level definitions plus the two nested diagram-generator
children, 46 namespaced skills in total. skills/ is the plugin folder; the root
marketplace points at it.
claude plugin marketplace add luongnv89/skills
claude plugin install lstack@ls
8000
tack --scope userFor one session from a trusted local checkout, run
claude --plugin-dir /absolute/path/to/checkout/skills. Skills appear as
/lstack:code-review, /lstack:drawio-generator, and so on. Upgrading from lstack
0.4.0 or earlier: --plugin-dir now takes the checkout's skills/ folder, not the
repository root.
Some members edit files, commit and push, install tools, call APIs or open GitHub issues and PRs, each behind its own approval gates. skills/README.md lists what each skill runs, sends or fetches.
Updates, Codex builds, ZIP exports, external prerequisites and directory-portal caveats
are in the lstack runbook. Maintainers regenerate and check
the plugin metadata from packages/lstack/manifest.json with:
python3 scripts/build-lstack.py --write-plugin-metadata
python3 scripts/build-lstack.py --check-plugin-metadataPick one, several, or the whole catalog:
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill code-review
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill code-review --skill auto-push --skill test-coverage
npx skills@1.7.0 add https://github.com/luongnv89/skillsUse agent-skill-manager (asm) for a single TUI/CLI across agents:
npm install -g agent-skill-manager
asm install github:luongnv89/skillsasm search # find by name or description
asm list # show installed skillsOther install methods
Remote (no clone)
curl -sSL https://raw.githubusercontent.com/luongnv89/skills/main/remote-install.sh | bashNon-interactive:
curl -sSL https://raw.githubusercontent.com/luongnv89/skills/main/remote-install.sh | bash -s -- \
--skills "code-review,auto-push" --tools "Claude Code" --scope globalClone + local
git clone https://github.com/luongnv89/skills.git
cd skills && bash install.shgraph TD
A[User request or trigger phrase] --> B[Agent matches skill name]
B --> C[Loads SKILL.md + references/]
C --> D[Follows exact steps + templates]
D --> E[Quality gates + artifacts]
E --> F[Report / plan / files / PR links]
A skill is a self-contained playbook: frontmatter metadata, instructions, optional scripts, and reference docs. The installer copies it to the right path for your agent.
| Feature | What you get |
|---|---|
| Standalone | Any mix installs cleanly; zero shared runtime |
| Versioned | Per-skill semver + catalog release history |
| Structured | Steps, templates, checklists, self-validation |
| Tool-agnostic | Same skill works in Claude Code, Cursor, Windsurf, Codex, Copilot |
| Scannable | Tables, diagrams, short outputs |
| Suite support | Multi-phase products (e.g. diagram-generator) with independent phases |
Every skill is standalone. Install one or many.
Use:
npx skills@1.7.0 add https://github.com/luongnv89/skills --skill <name>One skill to invoke per task. The task skills call the related skills for you. In the two web task skills, evidence is captured once, the live scan runs at most once, overlapping checks are skipped where the member supports skip-checks, and remaining overlaps are merged into one row per defect.
| Task | Invoke | It runs |
|---|---|---|
| Optimize a website/app design | design-optimizer | dont-make-me-think · ux-ax-review · viral-product-evaluator · website-agent-readiness, then frontend-design only if you opt in to apply fixes |
| Optimize for SEO, AI-bot or app-store search | search-optimizer | Web: seo-ai-optimizer · website-agent-readiness · viral-product-evaluator. App store: aso-marketing · viral-product-evaluator |
| Take an idea to a build plan | product-planner | idea-validator → prd-generator → (brand-name-checker) → tad-generator → tasks-generator, resuming from the furthest existing artifact |
| Review or improve code | code-review | One skill, four modes: review, perf, clean, cleanup |
| Draw a diagram | diagram-generator | Routes to drawio-generator or excalidraw-generator |
These other task areas don't have a task skill. Each skill in them stands alone, so invoke them in this order:
| Task | Skills, in order |
|---|---|
| Prepare an app store submission | aso-marketing → appstore-assets → appstore-review-checker |
| Harden and ship a repo | security-setup → devops-pipeline → ship (auto-push and cleanup-project for day-to-day work) |
| Open-source a project | oss-ready → doc-manager → landing-page-generator → github-pages-publish |
| Brand and launch | brand-name-checker → logo-designer → landing-page-generator → github-pages-publish → product-launch |
| Run an agent fleet | herdr-agent or rex-agent or tmux-agent-comms (alternative backends) · issue-work-loop (Herdr) · opencode-runner |
| Category | What it covers |
|---|---|
| Code Quality | Reviews, cleanup, testing, optimization, usability |
| Shipping | Auto push, repo cleanup, pipelines, releases, GitHub Pages publishing, security hardening |
| Product Planning | Validation, PRDs, architecture, tasks, naming, launch planning |
| Career Research | Tailored job discovery with employer/product due diligence |
| Frontend & Design | UIs, logos, diagrams, design optimization |
| Documentation | Docs gen, READMEs, SEO, OSS prep, agent config |
| App Store | ASO, screenshots and creative assets, review compliance |
| Tooling | CLIs, installers, local models, agent comms |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| code-review | 2.3.1 | high | Review or improve code — 4 modes: bugs/security, performance, clean-code audit, slop cleanup |
| codebase-modernizer | 1.5.0 | max | Whole-repo audit + phased, testable plan to modernize a stale or messy codebase |
| test-coverage | 1.6.0 | low | Target untested branches and edge cases |
| user-test | 1.1.0 | high | Launch an app, API or CLI and test every feature as a real user; log bugs, perf and UX issues with evidence; fix and verify simple ones; report with a plan |
| dont-make-me-think | 1.7.1 | medium | Usability review using Krug's principles |
| ux-ax-review | 1.3.0 | high | Evidence-backed human UX + AI/search AX audit and approval-gated improvement plan |
code-review has four modes — pick by intent or pass mode:<name>:
| I want to... | Mode | It... |
|---|---|---|
| Find bugs, security, or quality issues in a diff | review (default) |
reads + reports prioritized findings |
| Make code faster / fix perf bottlenecks | perf |
reads + reports performance fixes |
| Audit readability/standards vs the Clean Code cheat sheet | clean |
writes an audit report (CLEAN_CODE_AUDIT.md) |
| Actually apply cleanup / refactor out AI slop & cruft | cleanup |
writes code (8-subagent refactor) |
code-review or codebase-modernizer? Scope decides. code-review inspects a diff, a PR, or a
file set and reports findings. codebase-modernizer audits the whole repo across all ten
dimensions — including dependency and runtime currency, which nothing else here covers — and converts
the findings into a phased sprint plan with milestones. It is read-only: dependency upgrades become
planned tasks with migration steps, never a bulk npm update. Reach for it when returning to a
neglected project or untangling one that has drifted.
Adjacent skills: test-coverage (generate tests for untested branches) · user-test (run the app and test it as a user) · dont-make-me-think (usability/UX review).
| Skill | Version | Effort | What it does |
|---|---|---|---|
| auto-push | 1.2.0 | low | Commit message + stage + push with secret and size checks |
| cleanup-project | 1.1.0 | high | Review uncommitted changes, update ignore files, delete merged branches locally and on origin, end on clean main |
| devops-pipeline | 2.5.0 | medium | Pre-commit + lean, fast GitHub Actions quality gates |
| github-pages-publish | 1.1.0 | high | Publish or update a static site on GitHub Pages via GitHub Actions in any repo; URL returned only after a live check |
| security-setup | 1.6.1 | high | Local pre-commit secret scans, dep checks, static analysis, gated CI |
| ship | 3.1.0 | max | Autonomous release: version bumped everywhere, changelog, docs and landing page, tag, GitHub release, publish |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| product-planner | 1.2.0 | high | One run from idea to sprint tasks: idea-validator → prd-generator → tad-generator → tasks-generator, resuming from existing files |
| idea-validator | 1.10.1 | max | Market, feasibility, competitor checks for ideas |
| viral-product-evaluator | 1.8.1 | high | Score codebase + landing page vs 32 viral principles |
| product-launch | 1.2.0 | high | Pick launch sites from 50 (Product Hunt, HN, Reddit, directories) for a product, then write setup guides, posts and pitches after you approve the plan |
| brand-name-checker | 1.6.0 | max | Trademark, domain, social, registry conflicts |
| prd-generator | 1.6.1 | max | Structured PRD from idea or validate notes |
| tad-generator | 1.7.1 | max | Technical architecture document from PRD |
| tasks-generator | 1.6.1 | max | Sprint tasks and plan from PRD |
plan-to-issuesmoved to luongnv89/idd (idd#502), next to theissue-creatorit depends on:asm install https://github.com/luongnv89/idd --skill plan-to-issues, or the idd Claude Code plugin.
| Skill | Version | Effort | What it does |
|---|---|---|---|
| ai-job-scout | 1.2.1 | high | Rank open AI roles for a candidate; verify location and application links; assess each company's product, project, and technical interest with cited evidence; filterable HTML report for larger searches |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| frontend-design | 1.5.1 | high | Production UIs with usability-first approach |
| logo-designer | 1.5.1 | medium | 7 SVG logo variants from project context |
| diagram-generator | 1.5.0 | high | One entry point for diagrams — routes to draw.io XML or Excalidraw JSON |
| design-optimizer | 1.2.0 | high | One run to optimize a website/app design: usability, UX/AX, virality, agent readiness, opt-in fixes |
Diagram generator engines (install the umbrella or a single engine):
| Engine | Version | What it does |
|---|---|---|
| drawio-generator | 1.5.1 | draw.io XML — precise, editable, C4, swimlanes |
| excalidraw-generator< 8000 /td> | 1.6.1 | Excalidraw JSON — hand-drawn, sketch, wireframes |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| doc-manager | 2.2.0 | medium | Generate/update docs to match code, cited to path:line, never invented |
| landing-page-generator | 1.8.1 | high | Landing pages: marketing copy from a brief, or a README-to-landing rewrite, with privacy, terms, changelog, and docs pages, built as HTML with frontend-design |
| search-optimizer | 1.2.0 | high | One run to optimize SEO, AI-bot and app-store search: web or store branch, one merged report |
| seo-ai-optimizer | 1.7.1 | high | Technical SEO + AI-bot directives |
| website-agent-readiness | 1.5.0 | high | Scan a live site for agent readiness, plan the gaps, file them as issues |
| oss-ready | 1.5.0 | low | Add OSS files and templates |
| agent-config | 2.3.0 | medium | AGENTS.md by default (CLAUDE.md on request), shadow-checked and evidence-pruned |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| aso-marketing | 1.5.1 | max | App Store + Google Play keyword and metadata optimization |
| appstore-assets | 1.3.0 | high | Upload-ready App Store screenshots and creative assets for iPhone, iPad and Mac, including looping header and search videos, from code or a landing page |
| appstore-review-checker | 1.7.0 | high | Pre-submission audit vs Apple guidelines |
| appstore-tracker | 1.0.6 | high | Daily downloads, proceeds, store funnel, reviews and ratings for every App Store app via the asc CLI, with trend signals, ranked actions and a dated HTML dashboard |
| Skill | Version | Effort | What it does |
|---|---|---|---|
| cli-builder | 1.4.0 | high | 5-step CLI tool builder with approval gates and btop-inspired output |
| ollama-optimizer | 1.4.0 | medium | Hardware-aware Ollama tuning |
| install-script-generator | 2.4.0 | high | Cross-platform install.sh with env detection |
| opencode-runner | 1.7.0 | medium | Delegate work to opencode free cloud models |
| herdr-agent | 3.3.1 | medium | Manage Herdr agent fleets: tile panes, message/wait/read, steer, help |
| rex-agent | 1.1.0 | medium | Manage Rex agent fleets: equal columns, prompt/wait/read via program status, steer, help |
| issue-work-loop | 1.7.0 | max | Resolve one GitHub issue via a Herdr implementer→reviewer loop until CLEAN |
| tmux-agent-comms | 2.5.0 | medium | Spawn, message, read CLI agents in tmux |
| dev-machine-setup | 0.11.0 | high | Gap-driven dev machine setup/tune-up across macOS, Linux, Windows |
Most skills are skills/<name>/. Multi-phase products live under a suite folder: umbrella at skills/<umbrella>/ + phases at skills/<umbrella>/<phase>/.
Current suite: diagram-generator (draw.io + Excalidraw engines behind one router). Install the umbrella or any child. Installers discover both levels.
Task orchestrators (design-optimizer, search-optimizer, product-planner) stay flat at skills/<name>/. Their members are ordinary top-level skills, and a member can belong to more than one task: viral-product-evaluator and website-agent-readiness serve both design and search. Each orchestrator checks for its members in a Dependency Preflight and tells you how to install any that are missing.
Mirror the layout for your own multi-skill products.
| Doc | Purpose |
|---|---|
| CONTRIBUTING.md | Contribute skills; structure and versioning |
| docs/guide-building-agent-skills.md | Authoring guide (plan → write → validate → distribute) |
| docs/brand_kit.md | Logo files, colors, typography |
| docs/DECISIONS.md | Doc ambiguity resolutions |
| docs/troubleshooting.md | Install/setup validation fixes |
| docs/archive/ | Historical work notes (not current product docs) |
| CHANGELOG.md | Release history |
| SECURITY.md | Vulnerability reporting |
| CODE_OF_CONDUCT.md | Community standards |
Do I need every skill?
No. Pick only what you need. All are independent.
Which agents work?
Any that load external skill files. Installer tools (install.sh:23): Claude Code, Cursor, Windsurf, GitHub Copilot, OpenAI Codex, OpenCode, Google Antigravity.
How do I make my own?
Follow CONTRIBUTING.md, docs/guide-building-agent-skills.md, or patterns from existing skills.
Do skills change my runtime code?
No. They only guide the agent during development.
Supported Tool Paths (`install.sh:218-321`)
| Tool | Global | Project |
|---|---|---|
| Claude Code | ~/.claude/skills/<skill>/ |
.claude/skills/<skill>/ |
| Cursor | ~/.agents/skills/<skill>/ + .cursor/rules/<skill>.mdc |
same |
| Windsurf | ~/.agents/skills/<skill>/ + .windsurf/rules/<skill>.md |
same |
| GitHub Copilot | ~/.agents/skills/<skill>/ + .github/instructions/<skill>.instructions.md |
same |
| OpenAI Codex | ~/.agents/skills/<skill>/ + ~/.codex/AGENTS.md |
same |
| OpenCode | ~/.agents/skills/<skill>/ |
same |
| Google Antigravity | ~/.agents/skills/<skill>/ |
same |
Project Structure
.
├── skills/ # Skill sources; also the lstack Claude plugin folder
│ ├── <name>/
│ │ ├── SKILL.md # Required: frontmatter + instructions
│ │ ├── references/ # Loaded on demand
│ │ ├── scripts/ # Optional helpers
│ │ └── agents/ # Optional subagent prompts
│ ├── <umbrella>/<child>/ # Suite children (install.sh:44-46)
│ ├── README.md # Plugin directory listing page
│ └── .claude-plugin/ # Generated plugin.json
├── meta/<name>/ # Non-runtime files, kept out of the 512-file plugin folder
│ ├── docs/README.md # Human-facing docs
│ ├── evals/evals.json # Trigger/behavior suites
│ └── tests/ # Tests for the skill's scripts
├── eval-fixtures/<name>/ # Eval input fixtures
├── packages/lstack/ # lstack manifest + runbook
├── scripts/ # Build, validation and eval runners
├── tests/ # lstack packaging tests
├── docs/ # GitHub Pages site, guides, brand assets
├── examples/ # Worked examples (logo-designer case study)
├── .claude-plugin/ # Root marketplace.json
├── .claude/agents/ # Maintainer subagents for this repo
└── install.sh / remote-install.sh
Acknowledgements
- frontend-design draws from Anthropic's patterns.
- Many skills follow the conventions established in the skill-creator lineage.
Website · Claude How-To · Blog