Skip to content

AI Labs Claude Skills - Setup Guide

Source: ailabs-393/ai-labs-claude-skills via skills.sh - 42.1K combined installs across 27 indexed listings GitHub: ailabs-393/ai-labs-claude-skills (458 stars; MIT at repo level; last push 2025-11-11; layout packages/skills/<name>/SKILL.md for 23 skills plus a nested packages/skills/document-skills/<name>/SKILL.md for docx, pdf, pptx and xlsx, all mirrored into a committed dist/skills/ build copy; 24 generated index.js entrypoints, none implemented) Category: Business Operations / Generalist Productivity Quality Tier: 🟡 Trusted - a broad, script-backed business suite with 42,078 installs where all 27 indexed listings resolve to a real skill directory, MIT at repo level and clean Trust Hub and Socket verdicts on four sampled skills, held below Production by a last push eleven months before this snapshot, no CI beyond a funding file, generated placeholder index.js entrypoints, and four bundled document skills that carry Anthropic's own proprietary terms instead of the repo license

AI Labs (publisher ailabs-393) ships one of the larger generalist skill collections on skills.sh: 27 skills that cover trip planning, personal finance, sales reporting, investor decks, SEO audits, fiction continuity, Docker and CI/CD scaffolding, and the four Anthropic document skills. It is not a narrow toolkit. It is a small business-operations department written as markdown workflows, with Python scripts behind the ones that touch files.

The shape is consistent across the suite. Each skill directory holds a SKILL.md that opens with a short interview or clarification step (preferences, constraints, style), then a numbered workflow, then calls into scripts/ for anything deterministic. Skills that need memory persist JSON under ~/.claude/. That consistency is why the suite installs well and why it is worth reading before installing: the same front-loaded-question pattern explains most of its behaviour, and the same dist/ mirror explains why a naive file count double-counts it.

Installation

# The whole suite, 27 skills in one install
npx skills add ailabs-393/ai-labs-claude-skills

# The published npm package, Node >= 18 (README badge)
npm i ai-labs-claude-skills

The README's own build path, if you work from a clone of the repo rather than the npm package:

npm install
npm run build
npm run gen:pkg
npm run gen:index

npm install triggers the postinstall hook, which runs install-skills.mjs and copies dist/skills into <your project>/.claude/skills. Several skills call Python, so install their dependencies first: pip install -r packages/skills/data-analyst/requirements.txt covers pandas, numpy, scikit-learn, plotly, dash and dash-bootstrap-components; the document and deck skills need python-pptx, pypdf, lxml and defusedxml.

What It Provides

Skill Installs Use For
travel-planner 4,079 Trip planning with a stored preference profile: day-by-day itinerary, budget breakdown, packing list and cultural notes generated as JSON from plan_generator.py
research-paper-writer 3,271 IEEE and ACM structured academic papers: section scaffolding, citation handling and scholarly phrasing for a stated topic and length
personal-assistant 2,705 Task, schedule and habit tracking against a persistent JSON profile in ~/.claude/personal_assistant, with retention rules for stale context
finance-manager 2,482 Transaction extraction from PDF or CSV, spending analysis, savings rate, and an HTML financial report with charts
business-analytics-reporter 2,221 CSV sales and revenue analysis that names weak areas, interprets the statistics and writes improvement strategies
script-writer 1,894 YouTube scripts from a stored style profile: hooks, retention structure, tone and length for a given topic
nutritional-specialist 1,865 Meal plans, recipes and dietary advice filtered through stored allergies, restrictions and goals
pitch-deck 1,711 Ten-slide investor or sales deck written as a real .pptx through create_pitch_deck.py
startup-validator 1,562 Market, competitor and problem-solution validation, backed by the agent's own web research
codebase-documenter 1,513 README, architecture, API and code-comment documentation for an existing codebase, with beginner-first templates
social-media-generator 1,487 Per-platform posts for X, Instagram, LinkedIn and Facebook, written into an organised folder with meaningful filenames
storyboard-manager 1,423 Fiction continuity: character development, plot structure, timeline tracking and consistency checks over a storyboard folder
business-document-generator 1,403 Proposals, business plans and annual budgets filled from bundled PDF templates via generate_document.py
csv-data-visualizer 1,331 Plotly charts, statistical data profiling and multi-plot dashboards for a CSV file
seo-optimizer 1,276 HTML and CSS SEO audit: meta tags, headings, schema markup, plus sitemap generation from a site directory
data-analyst 1,173 Missing-value detection and statistical imputation, then a Plotly Dash dashboard over the cleaned data
docker-containerization 1,125 Dockerfile and compose generation for Next.js and Node, with docker-build.sh, docker-run.sh, docker-push.sh and docker-cleanup.sh
brand-analyzer 1,106 Brand archetype and identity analysis that outputs a written brand guidelines document

The remaining 9 indexed listings range from 890 to 987 installs: resume-manager (987, one-page PDF resumes built from a stored experience database), tech-debt-analyzer (985, code smell and dependency reports), cicd-pipeline-generator (980, GitHub Actions, GitLab CI and CircleCI configs), frontend-enhancer (968, Next.js styling, palettes and layout patterns), docx (922), pptx (914), pdf (903) and xlsx (890), which are the four Anthropic document skills carried in the nested document-skills folder, and test-specialist (902, unit, integration and end-to-end test authoring for JavaScript and TypeScript).

Why This Matters for Hermes Agents

Most of this suite is work a Hermes session already attempts ad hoc: a sales CSV turned into findings, a deck outline turned into slides, a site audited for meta tags, a repo documented. What the skills add is a fixed question order and a fixed output shape. The front matter of each SKILL.md states when to invoke, then the workflow states what to ask before writing anything. Loading one gives a session a specification instead of a blank page, which is the difference between a plausible report and a repeatable one.

The second gain is the deterministic half. finance-manager ships extract_pdf_data.py, analyze_finances.py and generate_report.py; csv-data-visualizer ships data_profile.py, visualize_csv.py and create_dashboard.py; pitch-deck and business-document-generator emit real .pptx and .pdf files. Those scripts make no network calls (no requests, urllib or curl anywhere in the bundled Python, shell or JS), so an agent can run them in a sandbox and check the artifact before showing it to anyone.

The third is the memory pattern. Five of the skills front-load an interview and write the answers to JSON under ~/.claude/: travel-planner and personal-assistant as per-skill directories, nutritional-specialist, resume-manager and script-writer as single files. Later runs read them back instead of re-asking. On a Hermes host that path belongs to a different tool, so you either repoint DB_DIR in the script or accept a stray directory, but the pattern itself is worth copying: collect preferences once, keep them small, expire what is stale. personal-assistant even ships explicit retention rules for discarding old context.

Finally, read it as a catalogue of prompts rather than as an SDK. Every npm index.js in packages/skills is a generated stub that logs the skill name and returns its input untouched, so nothing here runs through the package entrypoint. The value is SKILL.md plus scripts/, and the four document skills are Anthropic's own work redistributed under Anthropic terms, not original AI Labs content.

Usage

You say What happens
"Plan ten days in Japan for two, mid budget, no seafood" travel-planner collects or loads preferences, writes a day-by-day itinerary with a budget split and packing list through plan_generator.py
"Where is my money going this quarter" finance-manager extracts transactions from the statement PDF or CSV, computes category totals and savings rate, then writes an HTML report with charts
"Turn this sales export into a board update" business-analytics-reporter runs analyze_business_data.py, names the weak segments and drafts strategies tied to the numbers
"Build me an investor deck for this idea" startup-validator pressure-tests the market first, then pitch-deck writes the ten slides to a .pptx
"Write the README for this repo" codebase-documenter walks the tree, then produces a getting-started guide, architecture notes and API reference from its templates
"Audit our marketing site for SEO" seo-optimizer checks meta tags, heading order, schema and sitemaps with seo_analyzer.py and generate_sitemap.py, then lists fixes
"Containerise this Next.js app" docker-containerization writes a multi-stage Dockerfile and compose file, plus the build, run and push scripts

Verification

Confirm what the skills CLI installed:

npx skills list | grep -iE "travel-planner|finance-manager|pitch-deck"

Read a skill before trusting it. Note the filename case: this repo has no README.md, the file is ReadMe.md, so a lowercase probe returns 404. Every URL below returned 200 when checked with curl on Oct 26, 2026, and .../main/README.md returned 404 on the same date:

B=https://raw.githubusercontent.com/ailabs-393/ai-labs-claude-skills/main
curl -sL $B/ReadMe.md
curl -sL $B/LICENSE
curl -sL $B/packages/skills/travel-planner/SKILL.md
curl -sL $B/packages/skills/finance-manager/SKILL.md
curl -sL $B/packages/skills/seo-optimizer/SKILL.md
curl -sL $B/packages/skills/document-skills/docx/SKILL.md
curl -sL $B/install-skills.mjs

Prove the Python side imports before a task depends on it:

python3 -c "import pandas, numpy, plotly, dash; print('viz ok')"
python3 -c "import pptx, pypdf; print('docs ok')"

Security

skills.sh verdicts for sampled skills (verified Oct 26, 2026) - re-check the security pages on skills.sh before production use:

Skill Gen Agent Trust Hub Socket Snyk
travel-planner Pass Pass Warn
research-paper-writer Pass Pass Pass
finance-manager Pass Pass Pass
seo-optimizer Pass Pass Pass

The single non-Pass verdict is informative rather than alarming. Snyk's audit of travel-planner on Sep 15, 2026 rates it MEDIUM with one finding, W011 third-party content exposure at 0.10: the workflow runs web searches for destination information, so the agent ingests untrusted external pages. No sampled skill carries a malware, injection or exfiltration finding.

Read the write surface before installing, because none of it is declared. No SKILL.md in the repo carries an allowed-tools or permissions block, so whatever the host agent can do, these skills can do. Concretely: the memory skills create directories and JSON files under ~/.claude/; plan_generator.py, generate_report.py and the document scripts write output files wherever you point them; docker-push.sh runs docker push against whatever registry it is given, which publishes an image; and install-skills.mjs copies dist/skills into <project>/.claude/skills and then, 1.5 seconds later, recursively deletes its own package folder inside node_modules. Network access is not in the bundled code, it is in the instructions: travel-planner and startup-validator both tell the agent to search the web. Treat any skill that asks the agent to browse as untrusted-input handling, and keep the memory databases out of any directory you sync or publish.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations

  • The repo is a snapshot, not a maintained product. package.json sits at 2.0.0 and the last commit is 2025-11-11, eleven months before this sweep. .github holds only FUNDING.yml, and although package.json declares npm test -ws, the only test file in the tree is one pytest for the bundled pdf skill's bounding-box helper.
  • The root readme is ReadMe.md, not README.md. Any tooling or guide that probes the lowercase path gets a 404, and so does raw.githubusercontent.com/.../main/README.md.
  • The tree carries 54 SKILL.md files for 27 skills: packages/skills holds 27 and the committed dist/skills build copy holds the same 27 again. A raw file count therefore doubles the suite. The four document skills also sit one level deeper, under packages/skills/document-skills/, while skills.sh indexes them as flat ids (docx, pdf, pptx, xlsx).
  • License is mixed, not uniformly MIT. The repo LICENSE and root package.json say MIT, but document-skills/{docx,pdf,pptx,xlsx}/SKILL.md each declare license: Proprietary. LICENSE.txt has complete terms, and each bundled LICENSE.txt reads (c) 2025 Anthropic, PBC. All rights reserved with use governed by your Anthropic agreement. Redistributing or shipping those four under an MIT assumption is a licensing error.
  • The npm package is a delivery vehicle, not a library. All 24 index.js entrypoints are the stub that generate-index-files.js writes: they log the skill name and return the input unchanged. Do not plan integrations against them.
  • Persistence paths are hardcoded to Path.home() / ".claude" / <skill> in assistant_db.py, travel_db.py, script_db.py and resume_db.py, so on a Hermes host the data lands outside the Hermes profile and survives uninstall.
  • Only one of the four sampled skills carries a non-Pass verdict: travel-planner has the Snyk Warn for browsing, while research-paper-writer, finance-manager and seo-optimizer are Pass on all three scanners. The other 23 listings were not sampled, and startup-validator also instructs the agent to browse without a published audit of its own.
  • The same publisher ships ailabs-393/design-md-planner as a separate source with 32 installs. It is not part of this repo and is excluded from the 42,078.
  • Snapshot data, verified Oct 26, 2026: 42,078 combined installs across 27 indexed listings; 458 GitHub stars; MIT at repo level with the four document skills under Anthropic proprietary terms; last pushed 2025-11-11. Counts drift over time.

Pro Tips

  • Install the whole suite once, then delete the directories you will never invoke. 27 skills loading their descriptions into context costs more than the four or five you actually use, and what the CLI installs is plain folders, so dropping the rest is just a delete.
  • Repoint DB_DIR at your agent's own state directory before first use of travel-planner, personal-assistant, script-writer or resume-manager. Otherwise the profile JSON accumulates under ~/.claude/ and is invisible to every later session.
  • Treat postinstall as a write you did not ask for. It copies into <project>/.claude/skills and then removes its own node_modules folder; run npm i ai-labs-claude-skills --ignore-scripts if you only want the files.
  • For anything that produces a file (deck, report, resume, document), run the Python script directly and open the artifact. The agent's summary of a .pptx is not evidence that the .pptx exists or looks right.
  • Check the docx, pdf, pptx and xlsx licences separately before shipping output built with them. The repo says MIT, those four say Anthropic proprietary, and the second one governs.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing