Choosing the Right Lever: Prompt, Context, or Configuration

Where a fix lives decides whether it sticks

Having translated a reaction into a specific instruction, you face a choice that determines whether the fix holds: where to apply it. The same named change can go into the immediate prompt, into the context or knowledge base, or into a standing Project-level instruction, and the right choice depends on how often the change is needed. The Claude Certified Associate - Foundations (CCAO-F) exam treats this as an apply-level skill because getting the lever wrong is the single most common reason a correct fix fails to stick and the same correction keeps returning.

The judgement turns on a question about scope: is this a one-time adjustment, a piece of material every run consults, or a rule that should always apply. Each answer maps to a different lever. Misroute the fix, put a recurring rule in a one-off prompt, and you will re-apply it every cycle, which is exactly the correction treadmill the previous knowledge points warned about.

The three levers and what each is for

The prompt is the lever for a one-off. If the change matters only for this specific request, a wording tweak in the prompt is the right, lightweight home for it. Nothing about it should apply next time, so there is no reason to make it permanent. Using the prompt for genuinely one-time adjustments keeps the standing configuration lean and uncluttered.

The context or knowledge base is the lever for missing reference material that every run needs. If the fix is "the output was missing information the model should have had", a brand guide, the current product list, a policy document, that material belongs in the knowledge base where every run can consult it, not pasted into each prompt by hand. The standing instruction is the lever for a rule that should govern every future output: a tone constraint, a required section, a formatting standard. A rule is short and always-applies, which is exactly what a standing instruction is for. This three-way split is the same one formalised in the rule-reference-procedure placement test.

Matching the lever to how often the change is needed

The deciding question is frequency. Ask: will this same change be needed again, and if so, does every run need it or only some? A change needed once goes in the prompt. A change every run needs, if it is reference material, goes in the knowledge base; if it is a rule, goes in a standing instruction. That single question routes almost every fix correctly, and it connects directly to the promotion test in promoting a working fix into a standing instruction or Skill.

Two mismatches are worth naming because they are the exam's favourites. Putting a recurring rule in a one-off prompt means you re-add it every cycle, which is the correction that keeps recurring. Putting bulky reference material into a short standing instruction bloats every future prompt with content that belonged in the knowledge base, making every run heavier than it needs to be. The lever is not just where the fix technically works, it is where it works without creating a new problem.

What the CCAO-F exam trips candidates on

The first trap is fixing a rule that should apply to every future report by editing just the current prompt. It works for this one output, which makes it feel done, but next cycle the rule is gone and the same correction is needed again. The exam rewards recognising that an always-applies rule belongs in a standing instruction, so the fix survives every future run rather than evaporating with the conversation.

The second trap is loading reference material into a standing instruction meant for short rules. A standing instruction is for concise, always-on rules; stuffing a full brand guide or product catalogue into one bloats every prompt with content that should live in the knowledge base and be consulted as needed. The credited answer routes reference material to knowledge and keeps standing instructions lean. Both traps are the same mistake in mirror image: matching the lever to the content type and the frequency, not to whichever home is closest to hand.

Common misreadings to avoid

How this shows up on the exam

Domain 7 questions describe a fix and ask where it should be applied, or explain why a correction keeps recurring. The credited answer matches the lever to frequency and content type: one-off to the prompt, always-applies rule to a standing instruction, reference material to the knowledge base. When a scenario says "the same correction is needed every cycle," the diagnosis is almost always that a recurring rule was placed in a one-off prompt.

This knowledge point builds on turning a reaction into a specific instruction, which produces the named change to route, and it feeds promoting a working fix into a standing instruction or Skill, where the promotion decision is made. It shares its three-way logic with the rule-reference-procedure placement test in the optimization task statement. Choose the lever by frequency, and fixes stop recurring.