Evidence
Every run writes artifacts under .superloopy/evidence, so results can be checked after the chat.
01 / Evidence
Take control ofagent work
Superloopy runs agent work in a loop until proof exists: plan, act, evidence, gate.
Start loopScroll down
Evidence-first loops for Codex and Claude Code
Superloopy keeps agents moving inside a bounded loop, then lets evidence decide when the work is done.
Agents move fast. Superloopy makes done mean proven. Plans, commands, screenshots, and reports stay linked to the goal instead of disappearing into chat.
- 01Looped work
- 02Agent rhythm
- 03Proof gate
Skill lanes
Use loopy for focused work, research, frontend, and clone flows without changing the proof rule.
02 / Skill lanes
Visible progress
Plans, commands, screenshots, and reports stay linked to the goal instead of disappearing into chat.
03 / Visible progress
Final gate
Completion requires a named artifact and a clear pass, blocker, or next action.
04 / Final gate
Superloopy®
Agent work ✲ Loops ✲
Are all about proof
Bounded task loops
Read proof ⊕Evidence that survives the chat
Read proof ⊕A final report with blockers named
Read proof ⊕Proof
1 loop
One prompt becomes a bounded loop
Read proof ⊕4 steps
Running agent work has never been clearer
Read proof ⊕Proof
Done means proven.
Read proof ⊕01 / 03
Built for Codex and Claude Code workflows that need receipts
Focus on the task. Let the loop demand proof.
4 steps
Running agent work
Has never been clearer
01
Set the goal
Define the objective, success criteria, and evidence folder before the agent starts moving.
02
Act in loops
The agent keeps taking the next visible action, with each pass tied back to the goal.
03
Capture evidence
Commands, screenshots, reports, and audit files land where they can be opened and rerun.
04
Pass the gate
The final report separates finished work from blockers, with artifacts named in the open.
Install
Install once. Then type loopy.
Use the same evidence loop from Codex or Claude Code. The commands stay copyable without loading an app runtime.
Run in a terminal, then restart Codex and approve the hooks.
codex plugin marketplace add https://github.com/beefiker/superloopycodex plugin add superloopy@beefikerRun inside Claude Code, then approve the plugin hooks.
/plugin marketplace add beefiker/superloopy/plugin install superloopy@beefiker/reload-pluginsFAQ
01
What makes Superloopy different from a prompt?
Superloopy keeps quality tied to evidence. Goal criteria: the loop starts with a concrete objective and the artifacts needed to prove it. Evidence capture: commands, screenshots, reports, and changed files are written where they can be opened later. Gate check: Superloopy does not call work done unless the proof is named or the blocker is explicit.
02
Where does the evidence live?
Artifacts live in the project under .superloopy/evidence. That keeps the proof beside the work, not buried in chat history. The final report points back to the files, commands, screenshots, and notes that matter.
03
Does it work with Codex and Claude Code?
Yes. Superloopy is a command layer for Codex work and supports Claude Code workflows too. The same evidence rule applies across lanes.
04
What happens when the loop cannot finish?
Then the report says so. A loop can finish as passed, blocked, or incomplete, but it must leave a reason and the evidence it found. That is the point: no confident ending without proof. You get the next action instead of a tidy guess.