Skip to content
Loop Engineering
Loop Engineering Tools

Loop Engineering Readiness Score

Answer nine questions about your task. Score how ready it is for an autonomous loop — and see exactly what is missing before you let an agent run.

How the Readiness Score Works

01
Is the goal specific and outcome-based?
Goal clarity · 12 pts
02
Can success be verified by a command, test, score, artifact, or checklist?
Validation · 16 pts
03
Is there an independent checker that can reject bad or shortcut work?
Independent checker · 14 pts
04
Have you defined what the agent must not do?
Boundaries · 12 pts
05
Is there a hard stop rule?
Stop condition · 12 pts
06
Is there a budget or max iteration limit?
Budget limit · 10 pts
07
Is there a rollback or fallback plan?
Rollback strategy · 8 pts
08
Can the loop run in a worktree, sandbox, or isolated environment?
Sandbox isolation · 8 pts
09
Is human approval required before merge, deploy, delete, purchase, or external communication?
Human approval gate · 8 pts
56
/ 100
Manual supervision required
What is ready
Goal clarity is covered.
Independent checker is covered.
Sandbox isolation is covered.
What is missing
Add a machine-checkable validation step — the single most important part of a safe loop.
List forbidden actions so the agent cannot take unsafe shortcuts.
Define a hard stop: checks pass, max iterations, or repeated failures.
Cap iterations and spend so the loop cannot run away with cost.
Add a fallback: how to recover and what to summarize when blocked.
Require human approval before any high-risk, irreversible action.
Recommended next step

Define a hard stop: checks pass, max iterations, or repeated failures.

suggested-goal.txt
/goal [your outcome] — done only when:
  - validation passes (command/test/score)
  - an independent checker confirms the result
  - no forbidden actions were taken
Stop after N failed iterations or when the budget cap is hit.
Require human approval before merge / deploy / delete.
If blocked, summarize the blocker and recommend a human decision.

Loop Safety Criteria

Goal clarity12
Validation16
Independent checker14
Boundaries12
Stop condition12
Budget limit10
Rollback strategy8
Sandbox isolation8
Human approval gate8

Goodhart’s Law in Loop Engineering

If the validation metric becomes the target, the agent may optimize the metric instead of the real goal. Add boundaries that prevent shortcuts such as deleting tests, bypassing lint, or changing unrelated behavior. When stakes are high, add a separate checker so the loop cannot grade its own homework.