Tech Stack
Architecture
SKILL.md spec -> scripts/ directory -> Schema (inputs/outputs/credentials/cost) -> INDEX.md registry -> skills.sh publishing via GitHub. Each skill is self-contained and composable.Most developers use AI assistants for one-off tasks. I built a system where every task I complete makes the AI permanently smarter.
The Skill Architecture: Each skill is a self-contained directory: SKILL.md (instructions + context), scripts/ (executable code), and a typed Schema (inputs, outputs, credentials, cost). When Claude Code encounters a new task, it checks the skill index first — reuse before rebuild.
Composition Chains:
Skills compose into pipelines. full-client-onboarding chains: gmaps-leads -> classify-leads -> casualize-names -> instantly-campaigns -> welcome-email. One command goes from "I need leads in Austin" to a fully running email campaign.
The Self-Annealing Loop: When a skill fails, the error gets logged, the fix gets applied, and the SKILL.md gets updated. Next time, the same error can't happen. The system literally gets better every time it breaks.
Why This Matters: After 43 skills, I almost never start from scratch. Lead gen? There's a skill. Email campaign? Skill. Video editing? Skill. Browser automation? Skill. The compound effect means I'm 5x faster than when I started — and the gap keeps growing.
Published on skills.sh — install any skill with npx skills add aiagentwithdhruv/skills. Works with Claude Code, Cursor, GitHub Copilot, Windsurf, and 40+ other agents.
Want to build something like this?
I architect and deploy end-to-end AI systems — from MVP to revenue.
Let's Talk