// autonomous development control plane

Stop prompting.
Start looping.

Stop manually baby-sitting your IDE. soloopx. is an open directory of closed-loop recipes for coding agents. Provide a goal, lock in verification shell commands, and let Cursor or Claude Code loop autonomously until the exit checks pass.

active_session.sh
LOOP STATE: RUNNING
$
execute-loop --recipe=ship-pr-until-green
Iteration [1/10]
⚡ running feedback check: npm test
✖ Status Check: Failed (3 tests broken in src/app/loops/page.tsx)
🛠 Self-pacing: correcting implementation, editing broken files...
Iteration [2/10]
⚡ running feedback check: npm test
✔ Status Check: Passed (148 tests executed successfully)
EXIT CONDITION METAll checks green. Loop terminated. Ready to merge.
Works with AI Coding Tools
CursorClaude CodeCodexGemini CLIOpenCode
[ blueprint v1.0 ]

The Anatomy of a Loop

Most developers manually prompt coding agents step-by-step. Loops replace this slow hand-holding with a strict state-machine that operates autonomously until your criteria are fully satisfied.

Continuous Execution
ONE-SHOT

Standard Prompts

A prompt is a single ingredient. It provides goals but lacks checking gates. The agent acts once and stops, waiting for a human to check the output, write edits, and manually guide the next move.

Baby-sitting: you must manually inspect the output.
Vulnerable to hallucinations, drift, and fatigue.
No self-correction boundaries when tests fail.
CLOSED-LOOP

Curated Loop Recipes

A loop is a system. It wraps goals in automatic check commands and strict exit conditions. The agent executes, evaluates its work via actual terminal outputs, and self-corrects until the checks pass.

100% autonomous: the agent drives itself.
Infinite self-correction based on stdout error checking.
Saves hours of context-switching and terminal monitoring.

// loop_lifecycle_flow

[ STAGE 01 ]

Trigger

Booted manually by pasting our kickoff template, or automatically hooked into git events or save triggers.

[ STAGE 02 ]

Act

The agent iterates on your codebase—writing code, refactoring directories, and editing files to meet the goal.

[ STAGE 03 ]

Check

The agent runs a designated verification command (such as `npm test`) to read real status check results.

[ STAGE 04 ]

Exit

If the check fails, the agent restarts. When the exit check succeeds, the loop halts with absolute safety.

[ deployment instructions ]

Executing loops in your workspace.

Every loop recipe is built to run on top of standard modern CLIs. Zero custom server installs required.

1.

Open a loop detail page in our catalog and find the structured kickoff.

2.

Click Copy prompt to grab the complete instructions with checking gates.

3.

Paste the prompt directly into your coding chat environment (Cursor / Claude Code).

4.

For hook-based loops, check the install file JSON configs on the loop detail view.