- In short
- Promoting a working fix means applying the test "will this same correction be needed again" to decide whether a fix that worked once should be captured as a durable standing instruction or Skill rather than repeated by hand. A phrasing that reliably produces the right output is promoted to a standing instruction; a multi-step fix is promoted to a Skill so the whole procedure is captured, not just the outcome. A fix discovered and not captured will be rediscovered at the same cost every time the task repeats, by the same person or someone else.
Finding the fix is only half the job
The most expensive failure mode in troubleshooting is not failing to find a fix, it is finding one and then losing it. A correction discovered in one conversation, and left there, is gone the moment the chat ends. Next cycle, the same person or a teammate rediscovers it from scratch, paying the same cost all over again. The Claude Certified Associate - Foundations (CCAO-F) exam frames capture as an apply-level skill: once a fix works, decide whether it needs to become a durable part of the configuration, and if so, in what form.
The deciding tool is a single test: will this same correction be needed again? If the answer is yes, the fix does not belong in a one-off conversation; it belongs in configuration, where it applies automatically to every future run. This turns a repeated cost into a one-time cost, and it is the move that separates a workflow that improves from one that keeps solving the same problem forever.
- Promoting a working fix
- Capturing a fix that worked once as a durable configuration artifact rather than repeating it by hand. A reliable phrasing is promoted to a standing instruction; a multi-step fix is promoted to a Skill so the whole procedure is preserved. The promotion test: if the same correction would be needed again, it belongs in configuration.
The cost of a fix that is not captured
A fix left in a one-off conversation is not free to rediscover; it costs exactly what it cost the first time, multiplied by every person and every cycle that hits the same problem. The correction found on Monday is rediscovered next Monday, and by whoever else runs the task, each time at full price. Capturing the fix once converts that recurring cost into a single one, which is why capture is not an optional tidiness step but the point at which a fix actually starts paying off.
This is why "I'll remember it next time" is a trap. Human memory is unreliable across weeks and does not extend to teammates at all. A fix that lives only in someone's head, or in a closed conversation, is functionally lost to the workflow. The reliable homes are configuration artifacts, standing instructions and Skills, that persist independently of who is working and when, which is the distinction sharpened in memory versus configuration for persisting fixes.
Standing instruction versus Skill: matching the form to the fix
Once you decide to capture, the form depends on the shape of the fix. If the fix is a phrasing, a specific way of asking that reliably produces the right output, it promotes to a standing instruction. The instruction encodes the phrasing so every future run gets it automatically. This is the natural home for single, always-applies rules, the same lever discussed in choosing the right lever.
If the fix is a multi-step procedure, several actions in a particular order, it promotes to a Skill. A Skill captures the whole sequence, so the steps that made the procedure work are preserved, not just the final outcome. This distinction matters because squeezing a multi-step fix into a one-line standing instruction loses the steps: the instruction names the goal but not the method, and the method is what actually reproduced the result. Match the form to the fix, phrasing to instruction, procedure to Skill, and the capture holds.
What the CCAO-F exam trips candidates on
The first trap is leaving a fix that worked in a single chat, assuming it will be remembered next time without being captured anywhere. It feels resolved because this output is right, but nothing durable changed, so the next cycle starts from zero. The exam rewards recognising that a fix which will be needed again must be written into configuration, not trusted to memory or a closed conversation.
The second trap is promoting a multi-step fix as a single short instruction, which loses the steps that made the procedure work. A one-line instruction can name the desired outcome but cannot carry a sequence of actions, so the reproduced result becomes unreliable. The credited answer captures a multi-step fix as a Skill, preserving the procedure. Both traps come down to matching the capture to the fix: capture when it will recur, and capture in the form that keeps what made the fix work.
Worked example
An analyst discovers two things that fix her recurring outputs. One is a single sentence she adds to prompts that reliably makes Claude cite sources inline; it works every time. The other is a five-step process, load the data, compute the deltas, flag anomalies, format a table, write a summary, that she worked out to produce a clean monthly variance report. She is tempted to just remember both. What should she do with each?
Apply the promotion test first: will each correction be needed again? Both will, the inline-citation phrasing on every research prompt, the variance-report process every month, so both belong in configuration rather than in her memory or a closed chat. Choosing to just remember them is the first trap; next month, or the moment a colleague runs the task, the fix is gone and rediscovered at full cost.
Now match the form to the shape of each fix. The inline-citation fix is a single, reliable phrasing, so it promotes to a standing instruction. Encoded there, every future research prompt in the Project gets the citation behaviour automatically, with no retyping. That is the right, lightweight home for a one-sentence, always-applies rule.
The variance report is a five-step procedure, and here the second trap looms: compressing it into a one-line instruction like "produce a clean monthly variance report" would name the goal but drop the steps, load, compute deltas, flag anomalies, format, summarise, that actually make it work. So it promotes to a Skill, which captures the whole sequence in order. The result is that both fixes persist, each in the form that preserves what made it effective: the phrasing as an instruction, the procedure as a Skill.
Common misreadings to avoid
Misconception
If a fix worked in this conversation, it is resolved and will be available next time.
What's actually true
Misconception
A multi-step fix can be captured as a single short standing instruction naming the goal.
What's actually true
How this shows up on the exam
Domain 7 questions describe a fix that worked once and ask for the most durable way to keep it, or ask whether it should be a standing instruction or a Skill. The credited answer applies the promotion test, capture if it will recur, and matches the form to the fix, phrasing to instruction, procedure to Skill. Options that rely on remembering the fix or that flatten a procedure into one line are the distractors.
This knowledge point builds on choosing the right lever, which routes the fix, and it sets up memory versus configuration for persisting fixes, which explains why deliberate configuration beats relying on Memory. It also connects to consolidating steps and promoting patterns in the optimization task statement. Capture the fix in the right form, and a lucky round becomes a permanent improvement.
A user works out a reliable five-step procedure that produces a correctly formatted compliance report, and expects to run it monthly. What is the best way to capture it?
People also ask
When should a fix become a standing instruction?
When should a fix become a Skill?
What is the promotion test for a fix?
Watch and learn
Official Anthropic Academy lessons first, then hand-picked walkthroughs. Videos load only when you press play.
No videos curated for this concept yet
We are still curating the best official and community videos for this topic.
Official prep for this domain
Anthropic's own free prep module for this part of the syllabus, on the official prep course. Free with an Anthropic Academy sign-in.
References & primary sources
Master this concept with Archie
Practice it inside an adaptive study session. Archie, your Socratic AI tutor, tracks your mastery with Bayesian Knowledge Tracing and schedules the perfect next review.