CCAR-F practice questions
Anthropic · CCAR-F · 294 questions
Original practice questions for the Anthropic Claude Certified Architect – Foundations (CCAR-F) exam, covering agentic architecture and orchestration, tool design and MCP integration, Claude Code configuration and workflows, prompt engineering and structured output, and context management and reliability — framed in civic and municipal scenarios such as city AI pilots, public-health records, permitting chatbots, 311 assistants, courts / clerk workflows, and transit / utilities operations.
This course contains the use of artificial intelligence.
About the CCAR-F exam
- Exam fee
- $125 USD
- Time allowed
- 2 hours
- Questions
- 60
- Passing score
- 720 (scaled, 100-1,000)
- Languages
- English
- Format
- 60 items (multiple-choice and multiple-response, each item states how many responses to select); 4 scenarios drawn from a bank of 6; closed book, delivered through Pearson VUE either online proctored or at a test centre
Schedule this exam The certification this earns
Exam details published by the vendor, checked 20 August 2026. Vendors change fees and formats without notice — confirm on the vendor's own page before you book.
Practice Quizzes
Test your knowledge with standard 20-question practice sets.
Quiz 1
Quiz 2
Quiz 3
Quiz 4
Quiz 5
Quiz 6
Quiz 7
Quiz 8
Quiz 9
Quiz 10
Quiz 11
Quiz 12
Quiz 13
Quiz 14
Quiz 15
Browse by Domain
Study specific topics at your own pace.
Agentic Architecture & Orchestration · 81 questions
- A city 311 resident-services desk agent must keep calling MCP tools until the resident's request is fully resolved. Which stopreason should the loop treat as the signal to continue executing tools rather than ending the turn?
- A permitting chatbot calls a parcel-lookup MCP tool and receives zoning details. What must the agent do with that tool result before the next Claude turn?
- A transit rider-alert agent receives ambiguous requests that sometimes need schedule lookup, sometimes disruption status, and sometimes both. How should tool selection be designed?
- A water-utility dispute agent incorrectly ends its agentic loop as soon as assistant text appears, even when a pending tooluse block is present. What is the problem with that approach?
- A court-clerk assistant caps the agentic loop at five iterations as its primary stop rule and often leaves fee-refund checks unfinished. What should architects recognize about that design?
- A housing waitlist helper should keep working until Claude finishes assisting the resident. Which control-flow rule correctly implements that agentic loop?
- A 311 multi-issue agent looks up trash pickup, then needs a second tool for a related pothole ticket. Why must tool outputs re-enter the same conversation between iterations?
- A parks permit agent team proposes checking whether assistant text contains the word "done" to decide when the agentic loop is complete. What should the architect conclude?
- A city policy-research system uses one coordinator and several research subagents. How should inter-subagent communication be routed for observability and error handling?
- A county open-data research system spawns subagents for separate datasets. What should architects assume about each subagent's conversation context?
- A planning-staff coordinator handles both simple zoning lookups and complex multi-source synthesis. How should it choose subagents?
- A library research coordinator assigns several subagents to gather sources for a civic literacy report. What practice best minimizes duplicated source pulls?
- A housing-policy coordinator reviews synthesis output and notices coverage gaps on eviction-prevention programs. What should it do next?
- A transit-equity study agent decomposes corridor analysis into extremely narrow micro-tasks and later misses whole corridors. What risk does that illustrate?
- A public-health literature coordinator collects findings from multiple research subagents before writing for the board. What is the coordinator's primary role at that stage?
- An emergency-management after-action coordinator needs controlled information flow among damage-assessment, logistics, and communications subagents. Which communication pattern should it use?
- A municipal grants coordinator must spawn research subagents during a funding-scan workflow. What configuration is required for that invocation path?
- A permit-research subagent needs prior GIS findings to continue a parcel review. How should those findings be supplied?
- A city council briefing system configures a write-report subagent that must not call refund or payment tools. Which mechanism is appropriate?
- A school-district modernization team finished a shared baseline analysis in Claude Code and now wants to compare two refactoring approaches without losing that baseline. What should they use?
- A clerk e-filing synthesizer subagent must produce a unified summary from web-search and document-analysis specialists. What should the coordinator embed in the synthesizer prompt?
- A water-quality research pipeline must keep source URLs attributable while feeding excerpt text to a synthesis subagent. How should that information be passed?
- A transit planning coordinator needs corridor studies for three routes at once. How can it spawn that work efficiently in one model turn?
- A parks master-plan coordinator writes prompts for research subagents that may discover unexpected trail-usage patterns. How should those prompts be designed?
- A treasury refund agent must not call processrefund until getresident returns a verified resident ID. How should that prerequisite be enforced?
- A city 311 agent receives one resident message covering trash pickup, a pothole, and a billing dispute. How should the agent handle the multi-concern request?
- A housing caseworker must take over when an agent cannot finish a subsidy issue. What should the escalation packet include for a human without transcript access?
- A permitting fee-waiver path can adjust charges only after identity verification. Why is deterministic enforcement required here?
- A court-fee waiver agent must enforce prerequisite gates before filing adjustments. Which approach provides reliable enforcement when prompt guidance has non-zero failure risk?
- A utility shutoff-prevention agent escalates a hardship case to the human desk mid-process. What should the handoff list for staff?
- A public-health benefits agent must address eligibility and an address correction in one resident contact. How should those concerns be investigated?
- A solid-waste billing agent must verify the account before adjusting charges. Why is prompt text alone insufficient to enforce that order?
- A multi-MCP civic agent receives Unix timestamps from one system and ISO dates from another. How should those formats be normalized for reliable reasoning?
- A treasury agent must stop refund tool calls that exceed a policy threshold and send them to human escalation. Which mechanism provides that guarantee?
- A permitting agent must guarantee fee-cap compliance on every adjustment. Why choose hooks over prompt instructions?
- A transit fare-adjustment agent must prevent policy-violating tool calls at runtime. What should interception hooks do?
- A health-clinic agent consumes legacy EMR MCP tools that return numeric status codes. What should happen before the model reasons on those results?
- A housing subsidy agent has business rules that must never be probabilistically skipped. When should hooks be selected?
- A court payments agent pulls mixed date formats from clerk and finance MCP tools. How should cross-system formats be normalized?
- A 311 agent handles cash disbursements and sees a prompt-only rule never to refund over a threshold. Why is that insufficient?
- A city code-review bot must check each changed file locally and then assess cross-file integration. Which decomposition pattern fits best?
- A legacy permitting-system modernization keeps uncovering unknown dependencies during investigation. Which decomposition approach should the team use?
- A school-district test-coverage agent faces an open-ended codebase improvement task. How should it decompose the work?
- A water SCADA companion must review multiple services without diluting model attention. What should the architecture avoid?
- A GIS ETL remediation has stable, known steps that rarely change. When do fixed sequential pipelines beat adaptive decomposition?
- A transit mobile-app rewrite keeps revealing new constraints during investigation. What is the value of an adaptive investigation plan?
- A municipal website accessibility sweep must cover page-level issues and sitewide navigation. Which pattern should the agent follow?
- A county IT engineer paused an overnight Claude Code investigation and needs to continue the same conversation. Which approach resumes the named session?
- A city platform team finished a shared baseline analysis and wants to compare two testing strategies without losing that baseline. What should they do?
- After public-works code changes overnight, an engineer resumes a Claude Code session. What should they tell the resumed agent?
- After a county GIS schema migration leaves prior tool results stale, how should the platform team continue the investigation?
- A library digital team pauses Claude Code briefly overnight and most prior context remains valid. What should they do next morning?
- A fire-inspections refactor needs two data-model options explored from the same codebase analysis. Which session technique fits?
- Treasury MCP schemas rotated overnight during a batch-job investigation. Which continuation is most reliable?
- A housing authority casework assistant issues a request that invokes a long-running server-side tool. The response comes back with stopreason set to pauseturn and no final answer. What does this signal, and how should the caller resume?
- A parks and recreation assistant defines client-side tools for facility lookup and program registration. When the application returns tool output to the model, what must each toolresult block carry?
- A clerk-of-courts docket assistant executes a records-lookup tool that the model requested. In the Messages API, how are the results of that tool returned to the model on the next request?
- An elections office assistant produces one assistant turn containing three tooluse blocks: a precinct lookup, a ballot-status check, and a polling-place wait-time query. The application executes all three. How should the results be returned to the model?
- A water utility assistant receives an assistant turn containing two tooluse blocks: a meter-reading lookup and an outage-map query. The team decides the outage-map call is unnecessary, so they return a toolresult only for the meter reading and omit the other. The next API request fails. What is the correct explanation?
- A public-health records team delegates a records-cleanup analysis to a subagent after the parent agent has already read several schema files and run half a dozen tool calls. What context does the subagent actually receive?
- A public library delegates an audit of its summer-reading program data to a subagent. The subagent reads twelve files, runs several searches, and reasons through the discrepancies it finds. What portion of that work is visible to the parent agent afterward?
- A sanitation department launches a subagent to redesign collection routes, but the specification is ambiguous about whether holiday routes are in scope. Given that a subagent cannot ask the user a clarifying question mid-task, what is the correct design?
- A code-enforcement office launches a subagent to review a backlog of forty open violation cases. Partway through, the subagent considers splitting the backlog and launching subagents of its own. What is the sound architectural guidance?
- A fire-dispatch after-action review needs a single search across one directory of incident logs - a job that takes a few seconds and returns a handful of matches. An engineer proposes delegating it to a subagent to keep the main context clean. What is the right call?
- A county IT team is refactoring an authentication module. The work is a sequence of steps, but every step edits the same three source files and later steps depend on naming decisions made in earlier ones. How should this work be structured?
- A school district orchestrator fans out five parallel subagents, each summarizing one set of board policies. Three return usable summaries; two fail partway through with tool errors. How should the orchestrator proceed?
- A city treasury office launches eight parallel subagents, each drafting one section of a quarterly financial report. The published report must present the sections in a fixed, mandated order. What must the merge step do?
- An emergency-management office runs six parallel subagents, each producing a written assessment for one hazard category. The team must ensure that concurrent writes cannot destroy one another's output. What is the correct design?
- A GIS team plans a four-way parallel fan-out for a parcel-data modernization job. Reviewing the plan, an architect notices that step three consumes the parcel-boundary schema that step two produces. What is the correct reading of this plan?
- A procurement office must guarantee that an agent can never execute a command against the live vendor-payment system, and that the attempt is stopped before anything runs. Which hook event provides that control?
- An animal-control intake agent keeps finishing its response before a mandatory checklist - microchip scan recorded, hold period set, owner notification queued - has been completed. The team wants completion to be gated automatically. What is the right mechanism?
- A permitting department wants every agent session to begin with the current fee schedule and the list of suspended permit types already in context, without relying on the operator to paste them. Which hook event is designed for this?
- A 311 assistant must attach the current service-outage list to every resident message, and must block any message containing a full Social Security number before the model can see it. Which hook event handles both requirements?
- A transit agency instruments its agent runs with hooks and needs to record a metric each time a delegated agent completes its assigned work. Which statement correctly describes the relevant event?
- A housing authority runs a three-phase pipeline: intake review, eligibility determination, then notice drafting. Each phase is a separate agent run. What is the sound way to pass work from one phase to the next?
- A parks department agent processes 400 facility-inspection records. On the first run it fails after roughly 260 records. How should the job be designed so that re-running it is safe?
- A courts clerk's office runs an autonomous agent that reconciles case records against a docket system. During testing the agent occasionally enters a loop, repeatedly re-querying and re-checking without converging. Which control most directly bounds this behavior?
- A housing-authority workflow uses Claude Code to draft casework letters, and the team wants every file write to a case folder to be automatically checked against a redaction policy before it is considered complete. Which mechanism fits this requirement?
- An elections office runs a worker subagent that reports "all 47 precinct tallies reconciled." A downstream step then surfaces a mismatch in one precinct. What should the supervising agent have done before accepting the worker's result?
- A resident-facing 311 assistant built on Claude Code needs to let a citizen pause a multi-turn request-tracking conversation and pick it up again the next day without losing prior context, and separately lets a supervisor branch off an alternate resolution path to test without touching the citizen's live thread. Which two features respectively address these needs?
- A parks-and-recreation department wants every JSON program schedule that Claude Code writes to disk to be automatically checked for valid syntax immediately after the write completes, with the agent blocked from moving on if the file fails to parse. Which mechanism should enforce this?
Tool Design & MCP Integration · 52 questions
- A 311 MCP suite exposes lookupcase and getticket with near-identical one-line descriptions, so Claude misroutes calls. What should architects fix first?
- A permits MCP search tool description omits input formats and edge cases, so Claude often picks the wrong search tool. What should the description include?
- A public-health portal has a vague analyzecontent tool that overlaps lab extraction. What redesign best reduces selection errors?
- A housing authority agent relies on one generic analyzedocument tool for extraction, summary, and claim checks. What interface change helps most?
- A transit system prompt says "always search" and overrides careful MCP tool descriptions. What should architects review?
- A court clerk MCP set uses minimal one-word descriptions across similar filing tools. Why does selection stay unreliable?
- A water-billing MCP exposes accountlookup and meterreadhistory. What should each description clarify?
- A parks reservations MCP often confuses group versus individual bookings. What description content reduces ambiguous selection?
- A permitting MCP returns only "Operation failed" with no metadata, so the agent cannot decide retry versus explain. What should errors include?
- A 311 validation error should not be retried in a loop. How should the tool mark it?
- A housing search subagent hits a transient timeout on one source. What error-handling pattern should it follow?
- A utility MCP must distinguish empty meter history from access denial. Which design is correct?
- A court e-filing MCP call fails because the clerk role lacks write rights. How should the tool respond?
- A health MCP source times out mid-fetch after some lab rows arrive. What should the structured error carry?
- A transit fare tool rejects a policy-violating adjustment. How should structured errors steer the agent?
- A solid-waste MCP returns the same generic failure string for every problem. Why is that harmful?
- A synthesis-only civic research agent is given 18 tools including web search and starts mis-calling search. What distribution fix is best?
- A municipal report-writer subagent keeps calling web search instead of synthesizing provided findings. What tool-set change helps?
- A parcel enrichment coordinator must run extractparcelmetadata before any enrichment tools. How can it enforce that first call?
- A clerk extraction workflow must produce a schema tool call rather than free-form chat. Which toolchoice setting fits?
- A housing agent previously used unconstrained fetchurl for policy PDFs. What safer tool design validates civic document sources?
- A civic policy synthesizer often needs quick fact checks but must not get full open-data search rights. What tool-distribution pattern fits?
- A transit ops agent sometimes answers from context and sometimes must call a named MCP tool. What should architects understand about toolchoice?
- A city engineering team wants shared Jira MCP servers for every clone without committing tokens. Where should the MCP config live?
- A developer experiments with a GIS MCP that teammates should not inherit. How should it be configured?
- A permits team finds Claude picking weak Grep guesses instead of the capable MCP search tool. What should they improve?
- County IT needs standard Jira ticket workflows for service desks. What MCP approach fits first?
- A library catalog agent wastes calls exploring collection schemas. How can MCP resources help?
- A transit monorepo configures several MCP servers. What should be true after connection-time discovery?
- Treasury has unique levy calculation workflows no community server covers. When is a custom MCP justified?
- A municipal legacy explorer must find all callers of calculatefee across packages. Which built-in tool fits best?
- A school-district agent must locate test files by name pattern regardless of folder layout. Which tool should it use?
- Edit fails because anchors in a shared utility module are not unique. What should the civic coding agent do?
- A water SCADA helper needs to understand a large legacy tree without loading everything. What built-in strategy fits?
- A court case-management refactor must find usages of wrapper-exported helpers. What search sequence works?
- A parks booking service needs a small, precise change in one handler with unique surrounding text. Which built-in fits?
- A 311 triage agent must always call the classifyrequest tool first, before any other tool, on every incoming resident message. Which mechanism most reliably enforces that on a given turn?
- A 311 intake tool takes a complainttype string parameter that residents' free text maps to loosely, causing the model to send inconsistent values like 'trash', 'garbage problem', and 'refuse issue' for the same category. What schema change fixes this?
- A water-billing adjustment tool accepts an adjustmentamount number with no schema constraint, and a malformed agent call once sent a negative meter reading that silently created an absurd billing credit. Where should this be caught?
- A parks department exposes five near-identical single-purpose tools: getfieldname, getfieldcapacity, getfieldsurface, getfieldhours, and getfieldcontact, each returning one attribute of the same reservation record. What redesign improves this?
- A housing-authority document-verification MCP tool sometimes fails because the applicant's uploaded file is corrupted, and sometimes fails because the verification service is temporarily down. Both currently return the same generic error object. How should the tool's structured error distinguish them for the agent?
- A treasury payment-processing tool can fail three ways: a network timeout, a validation error on the payer's account number, and an authorization failure because the calling agent lacks permission. Architecturally, how should the agent's response differ across these three?
- A fire-dispatch agent already knows the exact file path of a single incident report it needs to review in full, with no need to search for it. Which built-in tool is the right fit for retrieving its contents?
- A sanitation-department agent needs to run an existing, already-reviewed shell script that converts a batch of route CSVs into the format the routing system expects. No existing built-in file tool performs this conversion. What is the appropriate tool choice, and what safeguard should accompany it?
- A building-inspections search tool can match thousands of historical inspection records for a busy commercial corridor, and returning every field of every match would overwhelm the agent's context. What payload design fixes this?
- A permitting agent's createpermit tool occasionally gets called twice for the same request after a network retry, and each call currently creates a brand-new permit record, producing duplicates. How should the tool be designed to make retries safe?
- A public-health agent has access to a deletepatientrecord tool that permanently removes a record from the system. What tool design pattern is most appropriate to prevent an accidental or misinterpreted call from causing irreversible harm?
- A transit-ops citation-lookup tool can match tens of thousands of records for a busy route and currently has no way to request results in pages, so every query either times out or silently returns an arbitrary small slice. What schema addition fixes this?
- A county animal-control office exposes a lookupimpoundrecord tool to Claude. Its inputschema defines the properties animalid, sheltersite, and intakedate, and the description of animalid reads "required - the tag number of the impounded animal." The required array in the schema is empty. In practice Claude sometimes calls the tool with no animalid at all. What is the correct explanation of how a parameter is actually made mandatory?
- A city permitting office connects two MCP servers to the same Claude assistant. The legacy building-permit system's server and the new fire-inspection system's server each expose a tool named getstatus. Staff report that status lookups are returning results from the wrong system. What is the appropriate architectural fix?
- A city's 311 resident-service assistant uses two tools: searchservicerequests, which only queries the request database, and closeservicerequest, which marks a resident's case closed. The team wants lookups to run without interrupting the agent while case closures always pause for a human dispatcher. How should the tools be designed to support that policy?
- A transit authority is building an MCP server that exposes on-time-performance data to Claude. Roughly forty analysts across three offices will use it, it must authenticate each analyst against the agency's identity provider, and it needs to be patched centrally when the schedule feed changes. Which transport choice fits these requirements, and why?
Claude Code Configuration & Workflows · 60 questions
- A new hire on the city apps team misses coding standards that lived only in a senior's user-level CLAUDE.md. What went wrong?
- A permits monorepo needs package-specific standards without duplicating a giant root file. What CLAUDE.md technique helps?
- A transit repo's CLAUDE.md became an unmaintainable monolith covering testing, APIs, and deploy. What should the team do?
- County IT wants every teammate to receive the same Claude coding conventions automatically. Where should shared conventions live?
- A library team's Claude sessions behave inconsistently across days. How can they verify which memory files loaded?
- A school-district SIS app folder needs module-local Claude notes that should not flood the whole monorepo. What scoping fits?
- A water utility engineer has personal editor preferences and the team has mandatory coding standards. How should they be separated?
- A city team wants every clone to get the same /review checklist for pull requests. Where should the command live?
- A verbose codebase-map skill floods the main municipal chat with analysis noise. How should the skill be configured?
- A destructive-migration skill for a civic database must allow file writes but block shell execution. What frontmatter helps?
- A city deploy skill needs the service name, but engineers often invoke it with no arguments. What frontmatter helps?
- A county engineer wants a personal brainstorming skill that teammates should never inherit. Where should it live?
- A housing authority needs on-demand release notes and always-on coding standards. How should those be split?
- An engineer keeps a personal /scratch helper that must not ship in the city repo. What scope is appropriate?
- Terraform conventions for the water utility should load only when editing Terraform files. What configuration fits?
- Test conventions for a parks monorepo must apply to test files scattered across many directories. What path rule helps?
- Why prefer path-scoped .claude/rules/ over always-on subdirectory CLAUDE.md files for niche GIS conventions?
- The permits API package needs conventions that should apply when working under src/api. What should the team create?
- A civic monorepo's test conventions cut across many packages. Why choose path-specific rules over many subdirectory CLAUDE.md files?
- Projection rules for a GIS package should activate only while editing matching geo modules. What does that illustrate?
- A 45-file library migration on the city platform has unclear sequencing risks. What Claude Code mode should come first?
- A one-line null-check bug in a single municipal fee calculator is already well understood. Which approach fits?
- Discovery across a noisy legacy clerk codebase floods the main chat. How can Claude Code isolate that pass?
- A transit microservice split has several valid architecture options still under debate. What should the team start with?
- After the team agrees on a planned SIS migration sequence, what is the sensible next Claude Code step?
- A clear stack-trace hotfix points to one emergency-dispatch file with an obvious fix path. What should the engineer do?
- A multi-integration permits rewrite could waste weeks if the wrong approach is coded first. How does plan mode help?
- A civic data-migration transform keeps drifting until engineers attach two or three concrete input/output pairs. What lesson applies?
- A utility CSV normalizer should improve through failing cases rather than guesswork. What iterative pattern fits?
- Before building a cache for parcel lookups, the team wants Claude to surface invalidation tradeoffs. What pattern helps?
- Date-and-timezone bugs interact in a clerk scheduler, while lint nits are independent. How should issues be batched?
- Null parcel IDs keep escaping a assessor helper despite vague "handle edges" guidance. What should engineers supply?
- A prose-only brief for a housing ETL transform keeps flipping output formats. What is the most effective fix?
- A civic ETL team wants Claude to implement a mapper with strong behavioral coverage. What should they prepare first?
- City CI must run Claude Code without hanging on interactive prompts. Which flag pattern fits?
- City DevOps wants PR review automation to emit machine-parseable inline comments. Which Claude Code CI pattern fits?
- CI-generated tests for the permitting service keep missing fixture conventions. What should the team put in project CLAUDE.md?
- After Claude Code generates a parks-fee patch in CI, how should automated review be run?
- A housing-portal PR keeps getting the same Claude Code review comments on every re-run. What should re-reviews do?
- CI asks Claude Code to generate tests for a transit fare module that already has a suite. What context should be supplied?
- A parks-and-recreation engineering team has a root-level CLAUDE.md with general repository conventions and a service-specific CLAUDE.md inside packages/reservations/ with rules just for that booking service. When Claude Code works inside packages/reservations/, how do these two files combine?
- A permitting engineering team wants a coding rule that applies only to files under services/fee-calculator/ without cluttering the main CLAUDE.md that every contributor reads. What is the appropriate mechanism?
- An emergency-management skill runs a lengthy damage-assessment analysis that produces a lot of intermediate tool output the main conversation doesn't need to see, only the final summary. Which skill configuration keeps that noise out of the main context?
- A housing caseworker wants Claude Code to propose a sequence of file edits to a benefits-calculation script and get explicit sign-off before anything is changed, given the high stakes of miscalculating a resident's subsidy. Which mode fits?
- A water utility runs a nightly Claude Code job in its CI pipeline that reads SCADA export files and writes a summary of anomalous pump-pressure readings. The job runs headless with no operator at a terminal, so any permission prompt would hang the build until it times out. The platform team wants the run to be able to read files and run git diff, and nothing else, without weakening the checked-in settings for engineers who use Claude Code interactively on the same repository. Which approach fits this requirement?
- A county public-health department wraps Claude Code in a small service that summarizes de-identified immunization-clinic notes. The dashboard team wants to show progress in the browser as the run proceeds, so they need machine-readable events emitted incrementally over the life of the run rather than one payload delivered at the end. Which --output-format value produces that behavior?
- A public library system lets branch staff use Claude Code against the catalog-integration repository. Policy states that the production ILS deploy command must never be executed from an assistant session, no matter how the request is phrased or how reasonable the model's justification appears. The architect wants a control that refuses the command outright rather than surfacing a prompt a tired staffer might approve. What should be configured?
- A sanitation department's IT group deploys a managed enterprise policy settings file to every city-issued laptop, denying a set of network commands from Claude Code sessions. A route-optimization developer adds an allow entry for one of those commands to the project .claude/settings.json and another to their personal user settings, then reports that neither one takes effect. What is the correct explanation?
- A code-enforcement office maintains a custom slash command that drafts inspection-notice language and is run dozens of times a day. The team wants that one command to always execute on a specific Claude model, regardless of which model the operator has selected for the surrounding session. Which mechanism expresses this?
- A fire dispatch center has a custom slash command that produces an after-action summary for a single incident. Today the incident number is hard-coded and shift leads copy the file and edit it for each call, which has already caused one summary to be filed against the wrong incident. The architect wants one command definition that accepts the incident number when it is invoked. What should the command use?
- A county IT team keeps a /deploy-check slash command that is supposed to reason about the repository's current state before a release window. Staff report it often describes a branch state that is hours stale, because the command's text only tells the model to consider the working tree. The architect wants the actual git status output present in the prompt at the moment the command runs. What should the command contain?
- A school district's data engineer is asking Claude Code to update a transformation that reads the student-enrollment feed. The engineer's first attempts described the feed's column layout from memory, and the model produced code against fields that no longer exist. The authoritative schema lives in a file in the repository. What is the most reliable way to get that schema into the prompt?
- A city treasurer's office wrote a skill that reconciles daily receipting exports against the bank feed. The skill works correctly when a user invokes it by name, but during ordinary reconciliation conversations the model never reaches for it. Its frontmatter description reads simply Reconciliation helper. What is the most likely cause?
- An emergency-management office is packaging its incident-action-plan drafting procedure as a Claude Code skill so the whole team gets the same workflow. For the harness to discover the skill, where must the SKILL.md file be placed?
- A municipal GIS team's parcel-QA skill has grown to include a full coordinate-reference-system conversion table and the county's twelve-page attribute-domain dictionary. Both are needed occasionally but not on most runs, and the SKILL.md is now unwieldy. What is the recommended structure?
- A procurement office wants a guarantee that every time a solicitation document is written into the repository, a redaction linter runs against it. An analyst added a line to CLAUDE.md reading Always run the redaction linter after writing a solicitation file, and it works most of the time but has been skipped on several long sessions. The compliance lead needs it to be unconditional. What should the architect do?
- An animal-control team runs a one-week Claude Code batch that migrates legacy kennel records. For this batch only, every run must follow a temporary convention: dates are to be emitted in the legacy vendor's format. The team does not want this convention checked into the repository's rule file, where it would confuse developers after the migration ends. What is the appropriate mechanism?
- A permitting modernization effort keeps the intake web application and the fee-calculation service in two separate repositories. An architect working in the intake repository needs Claude Code to read the fee service's source while implementing a new plan-review fee, without moving or copying either repository. What is the appropriate way to do this?
- A 311 resident-services team is loosening permissions so routine sessions stop prompting for read-only git inspection. A developer proposes adding Bash to the allow list. The architect objects and wants the narrowest rule that still removes the prompts for git status. What should be added to the allow list?
- A transit agency's shared build pipeline began failing when Claude Code runs stalled on permission prompts. An engineer's fix was to add --dangerously-skip-permissions to the pipeline command, and the builds now pass. The security architect is reviewing the change. What is the correct assessment?
Prompt Engineering & Structured Output · 60 questions
- A city PR bot should flag comments only when claimed behavior contradicts the code. Which prompt approach improves precision?
- DevOps adds only-report-high-confidence wording, yet style nits still flood civic repos. What does that show?
- High false-positive style findings drown out real security issues in city DevOps review. What is the main risk?
- The permitting team wants clearer PR-review prompts. Which design best reduces noise?
- One PR-review category floods the clerk portal team with false positives while its prompt is being rewritten. What should they do meantime?
- Fee-logic review severity keeps drifting between reviewers. What prompt design stabilizes classification?
- Permit validators still over-flag after adding a vague be-conservative instruction. Why?
- Ambiguous 311 routing improves after adding a few worked examples showing why one tool was chosen. What technique is that?
- PR review comments for the water-billing service keep changing shape. How can few-shots help?
- Municipal code review false positives drop after examples contrast acceptable city patterns with real defects. What is being taught?
- Permit PDF extraction struggles when citations are inline in some packets and bibliographic in others. What helps?
- Public-health form extraction leaves empty fields on informal measurements and mixed table/narrative layouts. What should be added?
- Detailed instructions for clerk filing extracts still produce inconsistent fields until examples are added. What does that illustrate?
- After adding few-shots, the parcel-form extractor handles novel layouts it never saw verbatim. What capability is that?
- A clerk filing extractor must return schema-compliant JSON reliably. Which approach is preferred?
- With toolchoice set to auto, Claude sometimes chats instead of extracting permit fields. What should the team do for mandatory extraction?
- Required schema fields for optional permit attributes cause fabricated values. How should the schema change?
- Violation categories for code enforcement keep expanding beyond a fixed list. Which enum pattern helps?
- Mixed document batches for the assessor's office need metadata before enrichment. Which toolchoice pattern fits?
- Strict JSON schemas stop malformed fee extracts, yet stated totals still disagree with line items. What limitation remains?
- Scanned utility bills arrive with messy dates and amounts. What should sit beside a strict extraction schema?
- A parcel extract fails schema checks. How should the retry be framed for correction?
- Parcel retries keep failing because the needed grantor name exists only in an external deed not in the packet. What should the architect conclude?
- Code-review false positives keep recurring for the water-billing service. What structured field helps analyze dismissals?
- Fee extracts pass schema checks but stated totals may disagree with line items. What self-correction pattern helps?
- tooluse already blocks invalid JSON shapes, yet an APN lands in the wrong field. What kind of error is that?
- Multi-page permit packets sometimes list conflicting square footage. What structured signal should the extract include?
- Some extract failures are format mismatches; others lack any source for a required field. How should handling differ?
- Overnight permit-packet extraction is latency-tolerant and cost-sensitive. Which API approach fits?
- Pre-merge CI review for the permitting service must finish before merge gates. Why avoid Message Batches there?
- A batch job for assessor PDFs needs mid-request live MCP parcel lookups. What limitation applies?
- Some oversized assessor packets fail in a batch after chunking. How should failures be recovered?
- Thousands of assessor PDFs are queued for extraction. What should happen before the full batch?
- Batches may take up to about 24 hours, but the clerk SLA is 30 hours end-to-end. What must submission cadence account for?
- Generated CI patches for fee logic need a second look. Which review design is stronger?
- A large multi-file civic PR overwhelms a single review pass. What architecture helps?
- Verification passes for permit extracts need to route human review wisely. What should accompany findings?
- Self-review in the fee-logic generator session keeps defending its own choices. What limitation is showing?
- Cross-file data-flow bugs in a housing-authority PR are missed by single-file comments. Where should that analysis live?
- Subtle authorization bugs in a clerk portal patch were missed by the generating session. What often works better?
- An emergency-management after-action report is drafted by one agent, and leadership is concerned that a single pass may miss factual inconsistencies against the incident logs. What pattern improves reliability here?
- A housing-authority casework review prompt originally said "extract the fields accurately." Extraction quality stayed inconsistent across caseworkers' scanned intake forms. Replacing that line with an explicit, itemized rubric -- e.g., "a household-size field is correct only if it matches a number appearing on the form; do not infer from unit type" -- produced measurably fewer disputed extractions. What does this demonstrate?
- A building-inspection assistant must review several code sections, cross-reference measurements, and only then emit a single structured pass/fail verdict as JSON. The team has Claude reason through the cross-references in an unstructured scratchpad section of its response, then emit the final verdict in a clearly delimited JSON block at the end. What is the benefit of separating scratchpad reasoning from the final structured answer this way?
- A parks-and-recreation facility booking form has a variable number of attendees per reservation, each with a name and an age bracket. Which schema design best represents this in the structured extraction output?
- A housing authority's eligibility team sends Claude a 40-page tenant handbook, three policy memos, and a caseworker's free-text notes in a single prompt, followed by instructions on how to summarize the applicant's situation. The model keeps treating sentences from the caseworker's notes as if they were commands, and occasionally answers a rhetorical question that appears inside a policy memo. Which prompt-construction change most directly fixes this?
- A parks department builds a small service that turns shelter-reservation requests into a JSON record for its booking system. The prompt asks for JSON only, but roughly one response in six begins with a line such as 'Here is the JSON you requested:' before the object, which breaks the downstream parser. Which technique most directly guarantees the response begins as JSON?
- A court clerk's office calls Claude to produce a single tagged block of case metadata and wants generation to halt as soon as the closing tag is emitted, so no trailing commentary is billed or returned. Which API parameter is designed for this?
- An elections office runs a nightly job that extracts the precinct number, ballot style, and machine serial from scanned poll-worker incident forms. Two runs over the identical batch produced records that differed in wording and, on a few forms, in which serial was picked. Which sampling setting best matches this workload?
- A water utility's outage assistant repeats the same 300-word block — its persona, tone rules, and escalation policy — at the top of every user message in a long conversation. Where does this content belong in the Messages API?
- A public-health team asks Claude to answer questions about a 60-page communicable-disease reporting manual. The current prompt states the question first, then pastes the manual after it. Answers are serviceable but frequently miss details that are plainly in the text. What ordering change should the team make first?
- A library system uses Claude to decide whether a donated book fits its collection policy, returning a JSON verdict plus a rationale. A reviewer notices the model often commits to 'accept' in the first field and then writes a rationale that partly argues against it. How should the output be restructured?
- A sanitation department's missed-pickup triage returns a JSON object that its dispatch system parses with a strict reader. The model reliably produces the six agreed fields but sometimes adds helpful extras such as 'confidencenote' or 'suggestedcrew', and the reader throws on unrecognized keys. Which schema change addresses this?
- A code-enforcement office classifies inspection findings into a violation category that feeds monthly reporting. With a free-text category field, the same condition arrives as 'Overgrown vegetation', 'tall grass', and 'weeds/overgrowth', so counts by category are unusable. What is the right fix in the output contract?
- A fire-dispatch quality-review pipeline extracts a structured incident record from call transcripts. When the transcript never states the caller's callback number, the model simply leaves the callbacknumber key out, and downstream code cannot distinguish 'not stated on the call' from 'the extractor failed'. What schema change resolves the ambiguity?
- A county IT group defines a structured-output schema for asset-retirement records that three separate applications will consume. Today the meaning of ambiguous fields such as dispositiondate is explained in a paragraph inside the prompt, and one team has already copied the schema into a new service without that paragraph. What practice should the group adopt?
- A school district's transportation office gets a structured output with five levels of nesting — route, then stops, then per-stop riders, then per-rider accommodations, then contact records. The model fills the top two levels reliably but the deeper objects come back with fields at inconsistent depths and occasional empty containers. Which change most likely restores reliable output?
- A city treasury office extracts payment amounts from vendor remittance letters into a structured record. With the amount declared as a string, the model returns values such as '$12,400.00', 'USD 12400', and '12,400', and the ledger import fails or misreads them. What is the correct schema-level fix?
- An emergency-management office's public-alert drafter has accumulated a prompt with eleven prohibitions: do not speculate, do not use jargon, do not exceed two paragraphs, do not mention road names, and so on. Output quality has gotten worse, not better, and the model sometimes produces text that is technically compliant but useless. What should the office do?
- A GIS team extracts parcel attributes from scanned deed records. When a deed does not state the lot dimensions, the model produces dimensions that are plausible for the neighborhood, and reviewers have caught several fabricated values already. Which prompt-and-schema instruction addresses this most directly?
- A procurement office's bid-summary prompt returns inconsistent formatting: some responses are markdown bullet lists and some are JSON. Reading the prompt end to end, an architect finds an early section that says 'summarize each bid as a short bulleted list' and a later section that says 'return a JSON object with one entry per bid'. What is the correct diagnosis and fix?
Context Management & Reliability · 41 questions
- A long 311 session summarizes aggressively and starts losing dollar amounts and case IDs. What should be preserved separately?
- Verbose work-order tool dumps are bloating the 311 agent context. What should happen before accumulation?
- Aggregated subagent research for a policy brief buries key findings in the middle of a long prompt. What placement helps?
- Downstream context budgets are tight after several research subagents. What should subagents return?
- A multi-issue housing session jumbles IDs, amounts, and statuses across topics. What context design helps?
- A 311 resident explicitly asks to speak with a human. What should the agent do?
- Which triggers should a municipal agent treat as appropriate escalation conditions?
- A city architect wants to route hard cases using sentiment scores and the model's self-reported confidence. What judgment applies?
- A clerk lookup returns multiple residents with similar names. What should the agent do?
- How should a utility replacement agent learn when to resolve a standard meter swap versus escalate exceptions?
- A GIS search subagent fails during a zoning research workflow. What should it return to the coordinator?
- A valid GIS parcel query returns no matching parcels. How should the subagent label the outcome?
- A permit-data subagent hits a transient timeout. What error-propagation pattern is appropriate?
- During multi-source civic synthesis, some topics lack support because sources failed. What should synthesis do?
- Which multi-agent error patterns should a municipal architect reject?
- A sprawling legacy student-information-system exploration keeps losing findings across long sessions. What helps?
- Finding every refund-caller path in a county billing codebase is extremely verbose. What should the main agent do?
- A city codebase agent crashes mid-exploration. How should recovery be designed?
- During a multi-hour municipal codebase tour, context fills with discovery noise. What Claude Code practice helps?
- Before spawning the next wave of exploration subagents, what should the coordinator do with the prior phase?
- A clerk extraction pipeline reports 97% overall accuracy but fails badly on handwritten hardship affidavits. What does this illustrate?
- How should a city monitor error rates even among high-confidence permit extractions?
- Before routing field-level confidence for municipal forms, what should the team do?
- Low-confidence or contradictory fields appear on a digital permit intake. What should happen first?
- When may a municipality safely reduce human review volume on automated extractions?
- Policy research subagents gather citations for a housing brief. What must synthesis preserve?
- Two credible transit reports disagree on ridership. What should synthesis do?
- Open-data findings for a parks briefing omit publication dates. What risk does that create, and what should travel with findings?
- How should a final civic research brief present certainty?
- When synthesizing mixed civic source types, how should financial tables and news articles be rendered?
- A transit-corridor investigation session has grown very long, and the team notices Claude Code responses becoming slower and less focused as irrelevant early exploration accumulates in context. What should the team consider doing?
- A water-quality research agent is given one very long combined document containing lab results, regulatory citations, and complaint logs, and it reliably cites content from the beginning and end but misses a key finding buried in the middle. What does this illustrate?
- A parks master-plan agent needs to work through messy intermediate reasoning about trail-usage data before producing a clean final recommendation, and the session is starting to run long. Which combination of practices helps?
- A public-health benefits research agent produces a summary of eligibility rules and lists several supporting claims, some pulled confidently from a verified state statute and others inferred loosely from an outdated blog post. What should the agent's output preserve so a human reviewer can calibrate trust appropriately?
- During a long housing-authority casework session, Claude Code's context window fills with dozens of prior tool outputs and file reads. The team notices that instructions given early in the session are now being followed less reliably than instructions given recently. What phenomenon does this describe, and what is a reasonable mitigation?
- In a multi-agent 911-dispatch data pipeline, a specialist subagent encounters an unrecoverable error partway through validating incident records and cannot complete its task. What should happen next in a well-designed multi-agent workflow?
- A court clerk's office feeds a 40-page hearing transcript into a single prompt asking Claude to extract every stipulated deadline into a structured JSON list. Deadlines mentioned on pages 15-25 are reliably missed, while ones near the start and end are captured correctly. What is the most likely cause, and what is a reasonable fix?
- A 311 request classifier returns structured JSON with a category field and a confidence field (0-1). The team adds a rule: any result with confidence < 0.6 is automatically routed to a human dispatcher instead of auto-filed. What is this design pattern primarily addressing?
- A coordinator agent spawns a subagent to extract structured fee data from scanned water-bill images, then immediately passes that subagent's raw output into a second subagent that calculates late-payment penalties, without checking the first output's shape. The second subagent silently computes penalties against a missing dueDate field as if it were zero. What does this scenario illustrate, and what should the coordinator do differently?
- A housing authority runs a long agentic session that reads forty case files, extracts each household's recertification date, and at the end writes a single consolidated waitlist report. Partway through, automatic compaction fires. The agent then produces a report covering only the last few cases and describes the earlier ones vaguely. What is the best explanation and remedy?
- During a long session, a parks-department agent read summerprogramroster.csv early on, then spent many turns on other work and went through a compaction. The user now asks it to update three campers' session assignments in that file. What should the agent do first?
These questions are original practice material and are NOT actual exam questions or brain-dump content. All vendor marks are trademarks of their respective owners. This site is not affiliated with, endorsed by, or sponsored by Anthropic.