I built a linter for Claude Code artifacts (skills, subagents, slash commands, .mcp.json configs, and plugins) and then pointed it at a pile of public repos to see what people actually ship. A few of the numbers surprised me enough that I wrote them up.The one that stuck: across 87 public subagents, 57% (50 of 87) declare no `tools` list at all. That reads like a safe default, but it's the...
The core insight: AI sessions fail not from bad models, but from missing structure. The DRS (Deployability Rating Score) gives you a single number (0-100) that answers "can I actually ship this?"
Key components:
* Contract freezing (no silent interface changes)
* Mock expiration (30-minute max)
* Scope limits (5 files, 200 LOC)
* Time-based convergence gatesIt's MIT licensed.
Curious what patterns you've seen in your AI coding sessions.
https://github.com/sgharlow/claude-code-recipes
I've been using Claude Code daily for about 6 months. After building the same prompts over and over, I started documenting them as "recipes" - structured prompts with context about when to use them and what output to expect.This repo has 100 recipes covering common knowledge work tasks: - Meeting notes → action items - Status reports - Performance reviews - Proposals and presentations ...