Practice — CCAO-F Domain 4: Workflow Integration and Solution Design (16%)
14 scenario-based MCQs. Answer key + explanations at the bottom.
Q1
A content marketer runs the same task every week: drafting social posts that must match the company's brand voice guide and reference the current product one-pager. Each week she starts a new conversation and re-uploads both documents before she can begin. Which change best streamlines this recurring workflow?
A. Keep one continuous conversation running so the documents remain in context week after week B. Ask Claude at the end of each session to memorize the brand guide for future conversations C. Create a Project with the brand guide and one-pager as knowledge files plus voice instructions D. Save the brand guide as an artifact and reference that artifact in each new conversation
Q2
An operations team configures a Project for vendor-contract review and uploads all 60 historical contracts, the full 300-page procurement manual, and every past review memo "so Claude has everything." Reviews soon slow down and Claude begins missing stated instructions. What is the best correction?
A. Switch the Project to Opus, whose greater capability can absorb the full knowledge set B. Curate the knowledge files down to the reference material each review actually needs C. Move the documents out of Project knowledge and upload them into each conversation D. Split the work across several conversations so the files only load when mentioned
Q3
A financial analyst spends a long conversation teaching Claude her team's quarterly variance-analysis method — definitions, thresholds, and preferred output format. Next quarter she wants Claude "already briefed" without repeating that work. What should she do?
A. Bookmark the conversation and resume the same thread when next quarter begins B. Ask Claude to remember the method so it can recall it in future conversations C. Re-run the same explanation prompts at the start of each quarterly conversation D. Persist the method into Project instructions that future conversations start from
Q4
Mid-conversation, a project manager finishes using Claude to refine a survey's question wording and now wants help planning an unrelated team offsite. The conversation is already very long. Which response to the context limitation fits best?
A. Restart: open a fresh conversation for the offsite planning work B. Summarize: compress the survey discussion and continue in the same thread C. Persist: move the survey discussion into a Project before continuing there D. Continue in the same thread but tell Claude to disregard everything above
Q5
A learning-and-development lead is producing a new-hire handbook chapter with Claude. She expects many rounds of revision and wants to compare versions as the chapter evolves. Which output format should she work in?
A. Inline responses that she copies into a separate document after each revision round B. An artifact, so the chapter is a standalone deliverable she can edit and version C. A structured table summarizing the chapter's key points for downstream reuse D. A fresh conversation for each draft so that every version is preserved separately
Q6
An executive assistant wants Claude to produce a Friday summary of the week's unanswered emails and the following week's meeting load. Today he pastes screenshots of his inbox and calendar into the chat. Which approach best grounds this workflow in live data?
A. Share his mailbox credentials in the chat so Claude can sign in each Friday B. Upload exported mailbox and calendar files each week instead of screenshots C. Ask Claude directly, since it can natively read a user's email and calendar D. Enable the Gmail and Calendar connectors, granting Claude access explicitly
Q7
A recruiter asks Claude to summarize a candidate-tracker spreadsheet "in our Drive," but Claude replies that it cannot find or access the file. IT set up the Drive connector months ago. What should the recruiter check first?
A. That the connector is still connected and access to that file has been granted B. That the conversation has not filled the context window, which blocks connectors C. That the chosen model is large enough, since smaller models cannot use connectors D. That the spreadsheet is short enough, since connectors only pass small documents
Q8
A revenue-operations manager wants Claude to answer team questions from live records in the company's proprietary in-house CRM, which has no existing connector. The solution must be sustainable for the whole team, not a one-off. What is the associate-appropriate move?
A. Export the CRM to CSV every morning and upload the file to a shared Project B. Write an API integration script within the chat to query the CRM directly C. Scope the requirement and hand it to the technical team to build a custom connector D. Ask Claude to browse to the CRM's internal URL and read whatever records each question requires
Q9
A support team's weekly incident recap from Claude keeps arriving in an inconsistent structure, and a teammate proposes "asking engineering to build us an API pipeline so formatting is enforced in code." The recap is produced in claude.ai from the same source files each week. What should the team try first?
A. Accept the proposal, since consistent formatting reliably requires programmatic control B. Configure a Project whose custom instructions fix the recap's structure and template C. Switch the recap to Opus, since higher-capability models format more consistently D. Produce the recap across separate conversations, one for each section of the report
Q10
A consultant has a folder of 40 client interview transcripts. Task 1: "help me think through which themes might matter." Task 2: "go through every transcript, tag each one with themes, and build a summary spreadsheet." Which pairing of surfaces fits the two tasks?
A. Chat for Task 1; Cowork for Task 2 B. Cowork for Task 1; chat for Task 2 C. Chat for both, uploading all transcripts into each conversation D. Cowork for both, since files are involved in each of the tasks
Q11
A customer-success manager triages inbound customer emails the same way every day: categorize, set priority, draft a reply in a set format. She wants Claude Cowork to run this specific procedure on demand without re-explaining it — while leaving her other, unrelated tasks unaffected. Which configuration fits?
A. Global instructions that describe the triage procedure in her standing context B. A project that bundles her recent inbox exports with the triage history C. A plugin packaging the procedure for installation across her whole team D. A skill encoding the triage procedure, invoked when that task comes up
Q12
An enablement lead has built a Cowork skill for drafting renewal-quote emails and wants to package it into a plugin for the entire sales organization right away. According to Cowork best practice, which step must happen first?
A. Convert the skill's steps into global instructions so every teammate inherits them B. Rebuild the skill inside a Project so that it carries its own knowledge files C. Run the skill on representative real tasks and review its outputs before packaging D. Have the technical team re-implement the skill as an API-based workflow instead
Q13
An operations analyst's monthly close runs as one giant prompt: gather figures from three uploaded reports, reconcile discrepancies, draft commentary, and format an executive summary. Results are erratic — some months the reconciliation step is skipped entirely. Which redesign best improves reliability?
A. Decompose the process into sequenced steps, reviewing each output before the next B. Rewrite the giant prompt with more forceful, explicit language about never skipping any of the steps C. Escalate to the developer team to automate the entire close through the API D. Move the three reports into a single artifact so Claude sees them side by side
Q14
A team lead designing a new Claude-assisted onboarding workflow begins by mapping every step of the current process, clarifying what she is actually trying to achieve, noting what Claude is good and bad at, and marking which steps stay human. In 4D terms, which competency is she exercising?
A. Description — communicating the task clearly enough for Claude to deliver it B. Discernment — evaluating the output critically instead of simply accepting it C. Diligence — staying accountable and transparent about the AI's involvement D. Delegation — problem and platform awareness, splitting work human vs AI
Answers
Q1: C. A recurring task that needs the same reference material every time is the canonical Project use case: persistent instructions plus knowledge files mean every new conversation starts "already briefed." One endless conversation (A) degrades as early material competes with new material. Asking Claude to memorize (B) fails because nothing persists between conversations unless deliberately carried over. Artifacts (D) are output deliverables, not persistent reference storage.
Q2: B. Project knowledge counts against the context window, so the guidance is curate, don't dump — oversized knowledge crowds out instructions and dilutes them. Switching models (A) is the classic wrong-layer distractor; the root cause is context load, not capability. Re-uploading per conversation (C) carries the same context cost with extra manual work. Splitting conversations (D) doesn't reduce what loads into each one.
Q3: D. Durable knowledge that must survive into future work should be persisted — moved into Project instructions or a document that new conversations start from. Resuming the old thread (A) drags a degraded, bloated context into the new quarter. Claude has no cross-conversation memory, so (B) does nothing. Re-explaining each quarter (C) works but is the exact repetition persist is designed to eliminate — it loses on repeatability.
Q4: A. The new topic needs nothing from the old one, so the cheapest correct move is a fresh conversation. Summarizing (B) is the runner-up but loses because it spends effort carrying survey material the offsite task will never use — summarize is for continuing the same work. Persist (C) applies when knowledge must be reused durably, which isn't indicated. Telling Claude to ignore prior turns (D) doesn't free the context; the old content still occupies working memory and can dilute instructions.
Q5: B. A document that will be revised repeatedly and compared across versions is exactly what artifacts are for: standalone, editable, versioned deliverables in their own panel. Copying inline output elsewhere (A) sacrifices in-place iteration and version history. A structured table (C) is the right format for data feeding downstream use, not a prose chapter. One conversation per draft (D) fragments the work and loses the revision thread.
Q6: D. Connectors are the mechanism for grounding Claude in live organizational data (Gmail, Calendar, Drive), and they follow a permission-first pattern — access granted explicitly. Weekly file exports (B) beat screenshots but remain manual and go stale, losing on the "live data" requirement. Claude has no native access to anyone's email or calendar (C) — a documented misconception. Sharing credentials in chat (A) substitutes a data-sensitivity violation for the sanctioned permission flow.
Q7: A. Connector access is granted explicitly, per surface — a connector that exists org-wide doesn't mean this user has granted it, or that its scope covers this file, so verifying connection and permissions is the first check. Context exhaustion (B) degrades answers but isn't what "cannot access the file" indicates. Connector availability isn't gated by model size (C), and connectors are not limited to short documents (D).
Q8: C. No existing connector plus a proprietary system means custom integration work — the hallmark CCAO-F move is to scope the requirement and hand it to Developers/Architects to build (e.g., a custom MCP connector). Daily CSV exports (A) are the tempting runner-up but lose on sustainability: manual, stale, and dependent on one person. Writing integration code in chat (B) is outside the associate surface. Browsing to an internal authenticated app (D) isn't a reliable, team-wide grounding mechanism.
Q9: B. This is the reverse-escalation trap: the format problem is solvable in-product, so the associate-correct move is to configure the Project — custom instructions that pin the recap's structure and template. Escalating to an API pipeline (A) hands off work the chat surface already handles; escalation is for genuinely technical needs. Switching to Opus (C) is wrong-layer — the issue is missing standing instructions, not capability. Per-section conversations (D) multiply the inconsistency instead of specifying the format once.
Q10: A. Task 1 is ask-and-answer thinking — chat's home turf. Task 2 is delegate-and-review: a multi-step job acting across a folder of files and producing a filled spreadsheet, which is what Cowork's plan–act–verify loop exists for. Cowork for both (D) is the runner-up but loses because Task 1 gains nothing from file-level agency — it's a conversation. Uploading 40 transcripts into chat for tagging (C) forces a long, degrading conversation to do a Cowork-shaped job.
Q11: D. A named, repeatable procedure invoked on demand is precisely what a Cowork skill is — "customer email triage" is the course's own example. Global instructions (A) are the runner-up but lose on scope: they apply to every task, so triage rules would bleed into unrelated work. A project (B) bundles files and history for an initiative; it doesn't encode a procedure. A plugin (C) is the later packaging step for distributing validated skills to a team, not the tool for one person's recurring task.
Q12: C. Cowork best practice: validate skills before plugin-izing — run the skill on representative real tasks and review outputs, because a flawed skill inside a plugin scales the flaw across everyone who trusts it implicitly. Global instructions (A) are the wrong scope for a task-specific procedure. Projects (B) don't validate anything. Re-implementing via API (D) escalates work that needs no technical build.
Q13: A. Decomposition — splitting one big request into sequenced smaller ones with review between steps — is the tested fix for a mega-prompt that intermittently drops steps; each checkpoint also gives the Description–Discernment loop somewhere to catch errors. More forceful wording (B) is the runner-up but loses because instructions in an overloaded prompt stay diluted regardless of tone. Escalating to the API (C) is premature when an in-product redesign exists. Artifacts (D) are outputs Claude generates, not input containers.
Q14: D. Deciding what work to give the AI at all — problem awareness ("what am I trying to achieve?"), platform awareness ("what is this AI good/bad at?"), and task delegation (splitting steps between human and model) — is the Delegation competency, the 4D pillar that maps to this exam domain. Description (A) comes later, when she phrases the actual requests. Discernment (B) applies when outputs exist to evaluate. Diligence (C) is about accountability and transparency, not scoping the handoff.