Stop prompting. Start looping.
$ trigger → act once → verify → stop on proof
Interactive
Watch one round close.
Not a chatbot demo. A loop is only real when the round ends on proof or a hard stop.
Simulated locally. No agent is calling your repo.
- Awaiting trigger…
Library
Work types, not prompt piles.
Pick a lane. The full wall has every loop when you want to sort and copy.
Survive contact with reality
Repo readiness, CI, test coverage, flake kills.
Open →Keep recurring work moving
Handoffs, toolchain health, production sweeps.
Open →Source → proof → ship
Pre-publish checks, social → insight, buyer talks.
Open →“Looks good” is not evidence
Adversarial review, browser streaks, prompt regressions.
Open →Captures that stay usable
Ingestion QA, research → artifact, stale memory cleanup.
Open →Invisible obligations, handled
Calendar promises, inbox triage, refund follow-ups.
Open →Agent pack
Claude Code can pick the right loop.
Hook one MCP server. Browse the catalog, shortlist for a goal, render a protocol with state files, stop conditions, and proof.
mkdir -p ~/.ai-loop-library
curl -fsSL https://ailooplibrary.com/mcp/server.py \
-o ~/.ai-loop-library/server.py
claude mcp add ai-loop-library -- \
python3 ~/.ai-loop-library/server.pyPrompts start work.
Loops finish it.
Bounded rounds. Verifiable exits. No infinite agent thrash.