building-agents

claude-skills

Agent skills by @shaharsha - 13 production-grade Agent Skills that work in Claude Code, claude.ai, Codex, Cursor, and any other harness that reads the SKILL.md format.

MIT licensed. Built day-to-day; battle-tested in real projects.

---

Install

Claude Code (recommended)

/plugin marketplace add shaharsha/claude-skills
/plugin install shaharsha-skills@shaharsha-skills

This installs all 13 skills as one plugin. Or pick a subset:

/plugin install documents-and-decks@shaharsha-skills    # gdoc-sync + gslides-sync + gsheets + presentation-generator
/plugin install brand-and-visuals@shaharsha-skills      # brand-system + brand-assets + image-generation
/plugin install engineering-decisions@shaharsha-skills  # tech-design-doc
/plugin install building-agents@shaharsha-skills        # prompt-engineer + writing-project-memory
/plugin install utilities@shaharsha-skills              # namecheap-domains + office-render + viewing-videos

Manual (any other harness)

git clone https://github.com/shaharsha/claude-skills.git ~/shaharsha-skills
ln -s ~/shaharsha-skills/skills/<skill-name> ~/.claude/skills/<skill-name>
# repeat per skill, or symlink the whole skills/ dir

For Codex, Cursor, Gemini CLI, OpenClaw, etc., follow the same shape - point your harness at skills/<name>/SKILL.md.

---

Skills

Documents & decks

gdoc-sync

Push a local .md to an existing Google Doc, then fix the four things Google's converter breaks: #anchor links arriving as broken URL strings, other.md#anchor cross-doc refs, inline images at 1500pt+ wrecking layout, and missing RTL for Hebrew/Arabic. One Python script, stdlib + google-auth.

gslides-sync

Sister to gdoc-sync for .pptx -> existing Google Slides. Rewrites broken slide-anchor and cross-presentation links into native pageObjectId references, scales oversized images, applies RTL per text shape. Same service-account setup works for both.

gsheets

Comprehensive Sheets API v4 CLI - the Sheets-shaped sibling to gdoc-sync and gslides-sync. Read / write / append / clear cells; create / rename / duplicate / delete tabs; freeze rows, set column widths, merge ranges; style headers (bold + colored bg + frozen + filter as one atomic batch); apply borders, banding (zebra stripes), and conditional formatting; sort and filter; or drop to a batch-update escape hatch for charts, pivots, data-validation dropdowns, protected ranges, and gradient color scales. Same service-account auth as the other Google skills.

presentation-generator

Generate 16:9 PDF + PPTX decks where every slide is a custom AI-rendered image - not a templated layout with stock photos. Style locks globally via a reference image; composition varies per slide (full-bleed photo, infographic, architecture flowchart, big-number callout, timeline, quote card, etc.). Research -> narrative arc (SCQA / Duarte / Kawasaki) -> style lock -> parallel generation at concurrency 4 -> QA -> assemble. ~$3-5 in image-API spend for a 10-slide deck.

Brand & visuals

brand-system

Author a production-grade brand book + design system in one shot: long-form BRAND.md (20 sections), printable BRAND.html rendered to BRAND.pdf via Chrome headless, plus tokens.css (Tailwind v4 @theme + light/dark :root) and tokens.json (W3C DTCG, consumable by Style Dictionary / Tokens Studio). An anti-template interview won't finalize without one invented proper noun, three falsifiable principles, three real don'ts, and a 150-word voice sample. WCAG 2.2 AA audited at authoring time.

brand-assets

The mechanical-pixel sibling to brand-system. Five pipelines that run locally in seconds: vectorize (split-by-color-mask + potrace, not one-shot vtracer which muddies palettes), finalize-svg (snap fills to exact brand hexes, normalize viewBox), rasterize SVG -> pristine PNG, icon-pack (one SVG -> favicon + apple-touch-icon + PWA pack with iOS-opaque background and PWA-maskable safe area), color-audit (histogram opaque pixels per hex; fails on >1% drift). Bash + Python stdlib.

image-generation