Install,
don’t write.

The problem

You ask your coding agent for a rate limiter. A real one — token-bucket, per-provider caps, respects Retry-After. Thirty seconds later it has produced 120 lines of plausible-looking Python. The structure is fine. The variable names are fine. There is a subtle off-by-one in the refill logic and it swallows 429s from one provider because the header casing is wrong. You will find the first bug in a week, the second in production.

The agent did this because writing from scratch was faster than the alternative. Package search, reading three READMEs, comparing forks, picking one, reading source to confirm — ten minutes of token-heavy navigation. Generation is cheaper. Generation also loses.

The fix

CustomClaw is a curated registry of ~40 vetted utilities for LLM and agent code. Single-file where possible. Stable slugs forever. One-command install. Machine-readable catalog so the agent can decide in one fetch.

npx customclaw-cli add rate-limit-handler

For agents

• GET /api/catalog — full JSON catalog, no auth
• GET /llms.txt — agent onboarding, paste into context
• GET /openapi.json — OpenAPI 3.1 spec
• POST /mcp — Streamable HTTP MCP endpoint

Or install the stdio MCP server:

claude mcp add customclaw -- npx -y customclaw-mcp

The bet

The registry wins when an agent about to write a JSON repair function from scratch has a cheaper, more reliable option one tool call away. Every slug referenced from a CLAUDE.md is a permanent foothold — the next agent reading that project’s config will check the registry too. That compounds.