↓ Skip to main content
Yonk-Labs
Yonk-Labs
  1. Projects/

ai-skills

·3 mins
Claude-Code Skills Agents Workflow Content-Generation
Table of Contents

ai-skills is a library built to stop the failure modes that show up once an AI coding session runs long enough: declaring victory on a task it silently redefined, generating content that reads like a vendor pitch, auditing code without ever running it, or shipping a white paper with a fabricated stat. It’s 34 skills organized by pipeline stage, 12 behavioral rules that auto-load into every session, 8 reviewer/validator agents, 7 lifecycle slash commands, and 12 pre-built personas so generated content has a real voice instead of a generic one — installed with one script and usable from any project on the machine.

The pipeline
#

Skills are organized by stage of work rather than as a flat list: Understand (research-base, reverse-engineer, market-intel) → Scope (mission-brief, project-compass, whats-next) → Build (Claude Code itself, alongside superpowers and spec-kit) → Audit (aat, prod-ready, ux-audit, dry-check, anti-vibe, secret-scan) → Test (demo-steps, browser-test, user-test, edge-cases, breakme) → Ship (prod-ready, security-audit, launch-pad). Content generation (gen-blog, gen-deck, gen-podcast, and friends) runs orthogonal to that lifecycle — it can fire while a feature is still being built or months after launch.

What’s distinct about it
#

  • Persona-driven content generation. Twelve pre-built voices (each with ElevenLabs TTS settings), and every content-generation skill runs its output through /not-an-ai to strip AI writing tells before it ships.
  • Journey-based testing, not just unit tests. /demo-steps maps real user workflows as YAML; /browser-test runs them in Playwright across desktop, tablet, and mobile with click counts and error capture; /user-test simulates five different skill levels actually installing and using the thing.
  • Multi-document audits. /aat, /prod-ready, /dry-check, and /ux-audit don’t produce a single report — they produce a Gap Analysis, Risk Assessment, Fix Plan, and a task index ready to drop into an issue tracker.
  • Shared research caching. /research-base runs once and every downstream skill consumes it, instead of five separate skills each re-running the same competitive research.
  • The Rethinking, before any rebuild classification. /tech-regret opens with “if I were building this today, knowing everything V1 taught me, what would I do differently?” before it classifies a single component as borrow, rebuild, or discard.

Install
#

git clone https://github.com/yonk-labs/yonk-ai-skills.git ~/yonk-apps/yonk-ai-skills
cd ~/yonk-apps/yonk-ai-skills
./install.sh

Installs to ~/.claude/; rules auto-load in every session after that, in every project on the machine. The installer never overwrites an existing file with different content unless you pass --overwrite.

Complements, not competitors
#

Built to layer with, not replace: spec-kit owns the spec-to-code pipeline; superpowers owns plan execution and TDD discipline; ai-skills owns everything around the code — understanding, scoping, auditing, testing, writing, launching, and rebuilding.

Links#