The Tools I Built to Make AI Coding Agents 10x

I built a toolchain that makes AI coding agents dramatically more useful — open-source skills, persistent memory, free-LLM routing, and per-domain executors. This post is the full set and what each piece enables.

The toolchain

Tool	What it enables
softco	Plan / code / test / deploy as a software firm
skill-builder	Scaffold and audit new skills
common-knowledge	Persistent memory across sessions
openrouter-free-infer	Free-LLM CLI + skill
latex-engineer	LaTeX compiles and debug
manim-coding-skill	STEM animations via ManimGL
Godot-Skill	Type-safe GDScript
clean-my-mac	Safe macOS cleanup
seo-master-skill	SEO / GEO / AEO playbook

The pattern

Skills compose. softco plans the work; skill-builder scaffolds new capabilities; domain skills execute. common-knowledge persists context so the next session resumes warm. openrouter-free-infer routes inference cheaply.

I covered the workflow in detail in How I Use Claude Code to Ship Faster and the stack in My AI Agent Skills Stack.

What makes it 10x

Compound effects. Each skill is small alone. Combined, they remove the boring 80% of every project.

Persistent context. The agent does not start every session cold.

Cheap inference for cheap tasks. Premium models for premium tasks; free for the rest.

Open source. Every skill is on github.com/shihabshahrier. Use them, improve them.

What I'd build next

A "task router" skill that picks the right domain skill automatically based on intent. Some manual gluing remains.

FAQ

Q: Is this Claude-specific? A: No — the Agent Skills standard is editor-agnostic.

Q: What's the biggest single multiplier? A: common-knowledge. Persistent context dramatically reduces the cold-start tax.

Q: Can I contribute? A: PRs welcome on every skill.

---

Written by Shihab Shahriar Antor. See my projects or hire me at Shahriar Labs.