A collection of Claude Code skills — reusable, model-invoked workflows that give coding agents domain expertise on demand.
Built and maintained by Joe Cook (Cook Coding) — full-stack developer and digital solutions architect. Turning ideas into production-ready digital products.
📍 Florida, USA · cookcoding.com · @joecookcoding · LinkedIn · X @thejoecook
A skill is a self-contained folder with a SKILL.md file. Its frontmatter description tells the agent when to use it; the body tells the agent how. Skills load lazily — they cost nothing until something in the conversation matches their trigger, then the full instructions (plus any bundled references, templates, scripts, and evals) are pulled into context.
The result: instead of one model improvising builder, reviewer, and skeptic all at once, you get focused, repeatable, battle-tested procedures that fire exactly when they're relevant.
Each skill here follows a consistent layout:
<skill-name>/
├── SKILL.md # frontmatter (name + trigger description) and the core workflow
├── references/ # deep-dive material loaded on demand
├── templates/ assets/ # reusable scaffolds the skill fills in
├── scripts/ # bundled executables (when the skill runs code)
└── evals/ # test cases that verify the skill triggers and performs correctly
| Skill | What it does |
|---|---|
| adversarial-review | A structured, multi-agent review workflow. Splits the work across narrow roles — Orchestrator, Investigator, Devil's Advocate, Impact Tracer, Fix Planner, Triage Lead, Final Decision — to catch what a single pass misses. Use for thorough reviews, red-teaming, ship/no-ship calls, and high-stakes engineering or business decisions, with explicit human-approval gates on risky actions. |
| no-glaze | Senior-engineer feedback mode where accuracy beats agreeableness. Leads with the strongest technical truth, bans flattery openers, labels confidence, pushes back on insecure or over-engineered directions, and separates facts from assumptions. The blunt register; pairs with adversarial-review (the pipeline). |
| codebase-exploration | An investigate-before-mutate discipline (adapted from NVIDIA's TensorRT-LLM exploration playbook) for any bug fix, feature, or refactor. Search the concept not the symbol, read callers before callees, find existing helpers before writing new ones, and use a three-pass file budget to avoid the "read 30 files, still lost" spiral. |
| optimize-claude-memory | Audits and restructures a repo's agent memory into an AGENTS.md-only layout: a lean root AGENTS.md, per-directory AGENTS.md files, procedures moved into skills, and no CLAUDE.md (Claude Code reads AGENTS.md itself). Cuts per-turn token usage and keeps the docs portable across Claude Code / Copilot / Cursor / Cody. Always produces a plan for approval before editing. |
| skill-optimizer | Audits and upgrades existing Claude Code skills against Anthropic's skill-authoring best practices — trims bloated descriptions without dropping triggers, splits monolith bodies into references, flags missing gotchas and dead handoffs, and reports drift between a skills repo and the installed copies. Complements optimize-claude-memory (which owns a repo's AGENTS.md tree). |
| zod-validation-patterns | Zod v4 schema patterns for writing, editing, or reviewing validation code — locking schemas to generated API types, cross-field rules, custom error messages, transforms, zodResolver + react-hook-form, and a v3 → v4 migration reference. |
| Skill | What it does |
|---|---|
| founder-validation-skills | A 5-step workflow taking a founder from "I have an idea" to a validated landing page and growth plan — without building the wrong thing. Customer-pain research → lead-magnet selection → copy → site build → funnel + 30/60/90 plan. Don't build until pain is validated. |
| profile-hero-redesign | A 5-step workflow that turns a clean-but-generic profile, personal-brand, or service-provider hero into a focused, premium, conversion-driven layout. Diagnosis → message hierarchy → visual composition → CTA conversion → responsive polish. |
| Skill | What it does |
|---|---|
| ui-craft-rules | 39 checkable UI/UX rules (hierarchy, component choice, button copy, radius, colour, dark mode, motion, tokens, AI-ready design systems) plus a 22-point review list of the tells AI-generated UI leaves: action buttons in tab bars, dropdowns for three options, "Are you sure?" dialogs, light cards in dark mode. Works in code or in Figma. Download the ready-to-upload zip: dist/ui-craft-rules.zip. |
| Skill | What it does |
|---|---|
| legal-draft | Drafts or updates legal text: terms of use, privacy policies, cookie notices, NDAs, an MSA, DPAs, a HIPAA BAA, an advisor agreement and an offer letter from General Legal's CC0 templates (bundled), plus short notices and disclaimers backed by a project's own terms. It states only what the code does, never renumbers a published clause, and ends with the decisions that belong to the owner or a lawyer. Not legal advice. It is drafting help, and anything a customer's money or health depends on needs a lawyer's review. |
| Skill | What it does |
|---|---|
| heic-to-jpeg | Converts iPhone HEIC/HEIF images to widely-compatible JPEG, preserving EXIF metadata and orientation. Handles single files or whole directories, skips already-converted files, and auto-installs pillow + pillow-heif if missing. |
| publix-clip-all | A browser-console script that clips every available Publix digital coupon. It clicks Clip coupon buttons in throttled batches, presses Load more until the list runs out, and logs progress as it goes. You paste it into DevTools in your own signed-in session. It never asks for credentials or calls private APIs. Not affiliated with Publix. |
| commit | Generates a Conventional Commits message from your staged/unstaged diff for review — correct type, scope, and a body that captures the why. Runs on Sonnet for the turn; never commits for you. |
| pull-request | Drafts a PR title + body from every commit on the branch since its base, grouped by theme rather than chronologically. Auto-detects the base branch and emits ready-to-run git push + gh pr create commands. |
These are skills for Claude Code. Once a skill is on your skills path, Claude invokes it automatically whenever the conversation matches its trigger description — you don't have to call it by name.
Clone the repo and drop the skills where Claude Code looks for them:
git clone https://github.com/joecookcoding/SKILLS.git
# User-level skills (available in every project)
Copy-Item -Recurse SKILLS\* "$env:USERPROFILE\.claude\skills\"# macOS / Linux equivalent
git clone https://github.com/joecookcoding/SKILLS.git
cp -r SKILLS/* ~/.claude/skills/Or copy individual skill folders into a project's .claude/skills/ directory to scope them to that repo.
- Automatically — just describe your task. "Give this PR a thorough review" loads
adversarial-review; "convert these HEIC photos" loadsheic-to-jpeg; "clip all my Publix coupons" loadspublix-clip-all. - Explicitly — reference the skill by name, e.g. "use codebase-exploration before you touch this."
Joe Cook — Cook Coding. Full-stack development across Python, Next.js, React, Node.js, and Azure, with a focus on AI-powered tools, scalable SaaS, and technology for churches and nonprofits.
Each skill is independent — open an issue or PR against the relevant folder. New skills should follow the layout above and include a clear, trigger-rich SKILL.md description plus evals/ cases that prove the skill fires when it should.