SKILLs

A collection of Claude Code Agent Skills.

A skill is a folder containing a SKILL.md file — YAML frontmatter (name + description) plus markdown instructions Claude follows when the skill's description matches what you're asking for.

Skills

Skills are grouped by whether they operate on a codebase (built for Claude Code, where they have access to your local files and git) or are general-purpose (work on documents/content and are portable to the Claude apps too).

Code skills

These read and act on a local project — best used in Claude Code.

qna — Answer, don't change

Read-only Q&A mode. Ask a question and Claude investigates and explains — proposing changes in its reply but not touching your files until you explicitly approve. Triggers on phrases like "just answer, don't change anything", "explain how X works", or "don't edit yet".

security-audit — OWASP Top 10 (2025) + hardening review

Audits a codebase for security weaknesses against the OWASP Top 10 (2025) plus seven hardening areas: secrets management, data encryption, input validation/injection, auth, dependency/supply-chain, error handling/logging, and configuration/transport hardening. On invocation it asks which area(s) to investigate (or all), then reports findings with severity, file:line, and remediation — read-only, no code changes until you ask. Triggers on "do a security review", "find vulnerabilities", "check for hardcoded secrets / SQL injection", etc.

gitignore — Understand the repo, then write .gitignore

Analyzes the codebase to detect its languages, frameworks, and tooling, then creates or updates .gitignore with the right entries — grouped and commented, merged with any existing rules (no duplicates, never clobbers lockfiles or .example templates). Also flags files that are already tracked and need git rm --cached. Triggers on "fix/generate the .gitignore", "what should I gitignore", "stop tracking node_modules/.env", etc.

tidy — Organize the project, don't break it

Cleans up a messy file tree: groups loose files into logical folders (images/, scripts/, styles/…), removes empty directories, and deletes genuine junk. It understands the project first, updates every reference when it moves a file (so <img src>/imports don't break), never deletes load-bearing files even when empty (__init__.py, .gitkeep, manifests…), and shows a before→after plan for approval before changing anything. Triggers on "clean up the project", "organize the files", "remove empty folders", etc.

pre-push — Is it safe to push right now?

A pre-flight safety gate. Inspects the actual git state — staged diff, tracked files, source, history — for secrets/API keys, .env/.venv being tracked, PII, leftover conflict/debug markers, and .gitignore gaps, then gives a clear ✅ ready / ⛔ blockers verdict. Read-only; surfaces git rm --cached and reminds you to rotate any secret already in history, and defers to gitignore and security-audit for fixes. Triggers on "is this ready to push?", "can I commit this?", "did I leave any secrets in?".

optimize — Performance report (measure, don't guess)

Analyzes the codebase and writes a performance report structured around five principles: do less work (algorithmic complexity), respect the memory hierarchy (data locality), measure first (profile + Amdahl's Law), exploit parallelism, and cut overhead. Findings are ranked by impact with file:line, expected speed/memory effect, effort, risk, and a measured-vs-suspected confidence tag — and it flags premature/cold-path tweaks as not worth it. Reports first; rewrites hot code only when you ask. Triggers on "optimize the program", "make it faster", "find bottlenecks", "why is this slow?".

Non-code skills

General-purpose skills that work on documents/content rather than a codebase. These are also portable to the Claude apps (web/desktop) — upload the skill folder as a ZIP via Customize → Skills.

pdf-summary — Full, page-aware PDF summaries

Analyzes a PDF in its entirety and produces a structured, chapter-by-chapter breakdown with minimal compression. Reads every page (in batches for long docs), cites where each point comes from (page/section), explains diagrams and figures and how they relate to the content, and keeps any instructions/procedures complete rather than summarized. Triggers on "summarize this PDF", "analyze this document", "break down this PDF", or pointing at a .pdf.

Install

Quick (one line)

Using Vercel Labs' skills CLI:

npx skills add POSTTTT/SKILLs

Then restart Claude Code — skills are loaded at startup, not dynamically.

Install only the skills you want

Each skill is an independent folder, so you can install just the one you need:

npx skills add POSTTTT/SKILLs --list           # see what's available
npx skills add POSTTTT/SKILLs --skill qna      # install just one

Or run npx skills add POSTTTT/SKILLs with no flag and pick from the interactive list.

Manual (no extra tooling)

Copy a skill into your personal skills folder (available in every project on your machine):

# macOS / Linux
cp -r .claude/skills/qna ~/.claude/skills/

# Windows (PowerShell)
Copy-Item -Recurse .claude\skills\qna $HOME\.claude\skills\

Project-only

Skills in .claude/skills/ are available automatically when Claude Code runs inside this repo — no install needed.

Restart Claude Code (or start a new session) and the skill is live.