MANDATE → LOOP → DONE

One mandate.
One loop. Done.

An agent without a mandate wanders. A mandate without a loop never finishes. Write the goal, say how you will verify it, set the stopping rule — then let the loop run.

1 · Write your mandate

Goal — what must be true when it is done? Verify — how will you check it? Stop — when must the loop stop, no matter what? Max loops

2 · Watch the loop run (demo — simulated in your browser)

Press “Run demo loop” to see one mandate cycle through the loop.

This demo is a simulation. It does not run a real agent — paste the template below into Claude Code, Codex, Cursor or any agent tool to run it for real.

3 · Copy the mandate template

Why most agent loops fail

No verification
The loop keeps “working” with nothing to check it against. Every loop needs a check that can pass or fail.

No stopping rule
Without a stop, a loop runs until it burns budget or makes mistakes unattended. Every loop must be able to stop itself.

No memory
Each run starts blank and repeats the last run's mistakes. One mandate should hand the next one what it learned.

Questions

Is this a real agent?
No. MandateLoop designs the mandate and demonstrates the loop. Your agent tool does the real work.

What is a “mandate”?
One delegated job: a trigger, a goal, a verification step, a stopping rule, and memory to carry forward. Five parts, no more.

Is my mandate uploaded anywhere?
No. Everything on this page runs in your browser.