nodumb-loop

nodumb-loop

Do not run the task through a mandatory pipeline. Find the earliest unclosed layer, apply one primary skill, and after a new result choose the route again. A normal route may consist of one skill or require none at all.

Main filter: if the skill's result cannot change the next action, it is not needed right now.

Choose the current layer

What is currently undefined	Primary skill	Sufficient result
For whom, why, what experience or product model is needed	ask-nodumb	A task is chosen or one question is named that truly changes the direction
The task is clear, but the method, scope, transfer of a fact, or the next debugging attempt is risky	nodumb	There is a recommended move, its boundary, an applicable fact, and a verifiable sign of success
The direction and scope are chosen, but the contract may break at the intersection of states, order, time, data, rights, or integrations	edge-hunt	The contract is fixed and only real risky scenarios with checks are selected
The transition contract is already clear, but it may be unclear to the person whether the action was accepted, whether it is in progress, and how it ended	system-feedback	For ambiguous transitions, it is defined what the person will see during waiting, success, and failure

If the task is already defined and local, execute it directly. This skill does not replace implementation, code research, or result verification.

Transition rules

1. Start not from the first row of the table, but from the earliest unresolved.
2. If the blocking question relates to the current behavior or the cause of the symptom and it can be safely resolved by available observation, first obtain the fact via nodumb. Do not replace the fact being verified with a question to a human or a choice of new behavior; a human decision is needed only when facts are unable to separate the branches.
3. Do not load the remaining skills in advance. Apply the next one only if its trigger has emerged as a result of the current work.
4. edge-hunt и system-feedback do not choose the product itself or the scope. If an inconsistent decision is discovered there, return to ask-nodumb or nodumb.
5. If an applicable fact is missing, go to nodumb. The Loop does not choose between local observation, an external source, and human participation — this is part of the facts mode nodumb.
6. After the first failed check, obtain an observation on the broken layer. Before the second edit without a new fact, go to debug mode nodumb.
7. After a successful check, apply changelog-discipline, only if its own contract is fulfilled, then finish the work. Do not run the remaining skills for the sake of completeness.

Check the route

Before acting, briefly answer:

• which layer is currently not closed;
• which skill is capable of changing the next action;
• which observable result will allow exiting it.

If there is no concrete answer to the second question — skip the skill. If a new fact has refuted the chosen route, return only to the affected layer, and do not start the entire cycle over again.