Loop Library

A curated, searchable directory of reusable "agent loops" — practical AI-agent prompts with explicit checks and stopping conditions — published by Forward Future (Matthew Berman). Launched 2026-06-18 with ~50 loops. Source: X/@MatthewBerman, 2026-06-18

What it is

Loop Library (signals.forwardfuture.ai/loop-library, canonical catalog at signals.forwardfuture.com/loop-library) is a public catalog of repeatable AI-agent workflows: copy-pasteable prompts packaged as "loops" with clear success checks and stopping conditions, organized into categories like engineering, evaluation, content, and operations. Users can search loops and submit their own. The live page now shows 70 loops, updated 2026-06-26, and the machine-readable API returns the same count. Source: Loop Library live page, catalog API, and catalog markdown, 2026-07-03

The launch partnered with here.now — "instant web hosting for agents," billed as the fastest way for an agent to publish websites and store files in the cloud (give an agent's output a URL). Source: here.now, via X/@MatthewBerman thread, 2026-06-18

The catalog is also agent-readable: catalog.json, catalog.md, catalog.txt, llms.txt, RSS, and an /agents/ surface are linked from the live page. The repository redirects to Forward-Future/loopy, which contains the installable loopy skill for finding, adapting, auditing, running, debriefing, and publishing bounded loops. Source: Loop Library live links and Forward-Future/loopy, 2026-07-03

Where it fits

A loop is the unit of repeatable agent work: a prompt + checks + stop condition. That maps directly onto Kevin's existing frame — Borrowed Intelligence (Drain It into Plans) (drain strong-model intelligence into reusable plans), The Eval Loop (Slop Is an Output Problem) (score every output against a benchmark, gate failures), and the Loop project itself. Loop Library is the community/catalog layer for that pattern; the value for Kevin is as a source of proven loop templates to adapt, not a runtime. Reference-only on Active Stack (What to Actually Use) unless a specific loop earns a place in his harness. Quality varies (community-submitted), so loops should pass his own eval gate before adoption.


Timeline