CCA-F start here · Domain 1 · Multi-agent mind maps

Multi-agent research mind maps

Everything the CCA-F exam asks about a coordinator and its subagents, on one page. It starts with which exam scenarios bring these questions. Then six mind maps, each in the form "what you see, where the fault is, the fix", with the rule to remember, the exam-guide line it comes from, and one practice question.

Domain 1 · 27%Domain 2 · 18%Domain 5 · 15%Guide 1.2, 1.3, 1.6, 2.2, 2.3, 5.3, 5.6About 30 minutes

Which scenarios ask about multi-agent systems?

The exam shows you 4 scenarios out of a bank of 6. One is built around a coordinator and subagents. Two others reuse parts of it:

#ScenarioDomainsMulti-agent questions?Maps to study
3Multi-Agent Research SystemD1, D2, D5Yes, the core one. Coordinator splits the work across search, analysis, synthesis and report subagents. Reports need citationsAll six
1Customer Support Resolution AgentD1, D2, D5Partly. Same Domain 1 and 2 ideas (tools per agent, structured errors), but one agent with hooks4, 5
4Developer Productivity with ClaudeD2, D3, D1A little. The Explore subagent keeps noisy searching out of the main context2
2Code Generation with Claude CodeD3, D5No—
5Claude Code for Continuous IntegrationD3, D4No—
6Structured Data ExtractionD4, D5No—
What this means for you. If the four are picked at random, scenario 3 is on your paper about 2 times in 3. Domain 1 is the biggest domain at 27%, about 16 of the 60 questions, and it comes up whether or not you get scenario 3.

Guide §5, scenario 3: “Multi-Agent Research System — coordinator delegates to search/analysis/synthesis/report subagents; cited reports. (D1, D2, D5)”

First, the picture: hub and spoke

Coordinator splits, routes, combines Searchweb_search, fetch_url Analysisread_document Synthesisverify_fact (scoped) Reportformat with citations Subagents never talk to each other. Each starts with an empty context: it knows only what the coordinator sends.

Most multi-agent questions come down to one thing: find which box the fault is in. If the question says the subagents “worked correctly”, the fault is usually in the coordinator.

The six maps at a glance

#QuestionUsual answerPractise
1The report misses whole topics or repeats workFix the coordinator's split of the workCoordinator split
2A subagent doesn't know what it needsPass the context explicitly in its prompt. It inherits nothingSubagent context
3Fixed steps or adaptive?Known steps: a fixed chain. Open-ended research: plan from what you findChain or decide
4Which tools does each agent get?Only the few its role needs, plus a scoped shared tool for frequent needsToo many tools
5A subagent failed. What goes back?A structured error with the partial results, never silence or a full stopTool failure
6Sources disagree or citations are lostKeep a claim-to-source map. Show both figures with their sources and datesSources

A yellow row in each map is the runner-up: the case where a different answer wins because one fact in the question changes.

Mind map 1: The report misses whole topics or repeats work

What you seeWhere the fault isFix
Every subagent did its part well, but whole areas are missing from the reportThe coordinator split the topic too narrowlyFix the coordinator's decomposition so it covers the whole question
Two subagents search the same groundTheir scopes overlapPartition the scope so each has a clear, separate area
A simple question runs through every subagentThe full pipeline always runsThe coordinator picks only the subagents the query needs
The first draft has gapsOne pass and doneRefinement loop: check the synthesis for gaps, send new tasks, synthesize again
One subagent was given the right area but covered it badlyThat subagent (its prompt or its tools), not the coordinatorFix that subagent
Rule: if subagents did their assigned work well and the result is still incomplete, the coordinator's split is wrong. The coordinator splits the work, routes it, and loops back for gaps.

Guide 1.2: “Risk: overly narrow decomposition → incomplete coverage of broad topics. Dynamically select subagents rather than always routing the full pipeline; partition scope to minimize duplication; iterative refinement loops (evaluate synthesis for gaps → re-delegate → re-synthesize).”

Example. Asked about “renewable energy policy in Europe”, the coordinator creates tasks for solar subsidies and wind tariffs. Each subagent's findings are accurate. The final report says nothing about hydro, storage or grid rules. What is the cause?
  1. The search subagent's queries are too narrow
  2. The coordinator split the topic too narrowly
  3. The synthesis subagent dropped some findings
  4. The report subagent's template has too few sections
B. The subagents did what they were given, and the missing topics were never assigned. A is wrong because the searches matched their tasks. C and D are wrong because the missing material was never found in the first place.

Mind map 2: A subagent doesn't know what it needs

What you seeWhere the fault isFix
The synthesis subagent ignores findings the coordinator already hasSubagents don't inherit the coordinator's historyPut the complete earlier findings in its prompt
Citations get lost between agentsContent and source details are mixed in free textA structured format that keeps content apart from metadata (URL, document name, page)
The coordinator can't start any subagent"Task" is missing from allowedToolsAdd "Task" to the coordinator's allowed tools
Independent subagents run one after another and it's slowOne Task call per turnEmit several Task calls in one response so they run in parallel
Step B needs step A's outputRunning them in parallel gives B nothing to work withRun them one after the other
Subagents follow the coordinator's step list too rigidlyThe coordinator wrote a procedureWrite goals and quality criteria, not steps
Rule: a subagent starts with an empty context. Everything it needs goes into its prompt explicitly, in a structured form that keeps the sources. Independent subagents go out together; dependent ones go one after another.

Guide 1.3: “Task tool spawns subagents; allowedTools must include "Task". Subagent context must be provided explicitly in the prompt (no automatic inheritance / shared memory). Spawn parallel subagents by emitting multiple Task calls in one response; write coordinator prompts as goals + quality criteria, not step-by-step procedures.”

Example. The synthesis subagent keeps producing reports without citations, even though the search subagent recorded a URL for every finding. What should change?
  1. Tell the synthesis subagent to “always cite sources”
  2. Pass the findings to synthesis as structured records, each with its claim, URL and page
  3. Give the synthesis subagent the web search tool so it can find sources again
  4. Merge search and synthesis into one agent
B. Synthesis can only cite what it receives, so send the sources with each claim. A works only if it has the URLs, which it doesn't. C repeats work and gives synthesis tools outside its role. D throws away the benefit of separate roles.

Mind map 3: Fixed steps or adaptive?

What you seeWhere the fault isFix
The steps are known and the same every time (review each file, then check across files)—A fixed chain of prompts (prompt chaining)
Open-ended research where each finding changes what to look at nextA fixed plan can't follow new leadsAdaptive decomposition: create subtasks from what is found
One big pass over many documents gives shallow, contradictory resultsAttention is spread too thinOne pass per document, then a separate combining pass
An adaptive agent is used for a routine job with the same steps every timeIt adds cost and unpredictability for no gainUse the fixed chain
Rule: when you know the steps in advance, chain them. When the next step depends on what you just found, let the coordinator plan as it goes.

Guide 1.6: “Fixed sequential pipelines (prompt chaining) vs dynamic adaptive decomposition based on intermediate findings. Prompt chaining: analyze each file individually, then a cross-file integration pass. Adaptive investigation plans generate subtasks from what's discovered.”

Example. Researchers ask why a drug's trial results differ between countries. Early findings point at dosing rules, which then point at local regulations nobody had planned to check. Which design fits?
  1. A fixed chain: search, analyse, synthesize, report
  2. A coordinator that creates new subtasks as findings come in
  3. One agent with all the tools and a long prompt
  4. Run the fixed chain twice and merge the results
B. Each finding changes what to look at next. A works for routine questions, but it can't follow the new lead. C overloads one agent. D repeats the same blind spots.

Mind map 4: Which tools does each agent get?

What you seeWhere the fault isFix
One agent has 18 tools and picks badlyToo many toolsSplit by role so each agent has 4 or 5 tools
The synthesis agent starts doing web searches itselfIt has tools outside its roleScope tools per agent (set them in each AgentDefinition)
Synthesis needs many quick fact checks, and each round trip through the coordinator slows things downEvery small check goes the long wayGive synthesis a scoped verify_fact tool for simple checks
The check is complex and needs real researchA small tool can't do itSend it through the coordinator to the search subagent
Someone suggests giving synthesis all the search tools “to be safe”Too much accessKeep least privilege. A scoped tool covers the common case
Rule: each agent gets only the tools its role needs. For a frequent simple need, add one narrow shared tool. Anything bigger still goes through the coordinator.

Guide 2.3: “Too many tools (18 vs 4-5) degrades selection reliability. Agents with off-specialization tools misuse them. Scoped access; limited cross-role tools for high-frequency needs.” Guide 1.3: “AgentDefinition: descriptions, system prompts, tool restrictions per subagent.”

Example. 85% of the synthesis agent's checks are simple lookups like “what year did X happen”. The other 15% need real research. Round trips through the coordinator add 40% to the run time. Best fix?
  1. Give synthesis all the search subagent's tools
  2. Give synthesis a scoped verify_fact tool for simple lookups, and keep sending complex checks through the coordinator
  3. Collect every check and run them all at the end
  4. Cache extra context ahead of time
B. It removes most round trips with the least access. A works, but it gives far more access than needed. C makes synthesis wait on checks it needs now. D guesses what will be needed.

Mind map 5: A subagent failed. What goes back to the coordinator?

What you seeWhere the fault isFix
The coordinator gets only “search unavailable” and can't decide what to doA generic status hides the detailsReturn a structured error: type of failure, the query tried, partial results, alternatives
A timeout comes back as an empty, successful resultThe error is hiddenReport it as a failure. An access failure is different from a real empty result
One failed search stops the whole research runIt ends everything on a single failureLet the coordinator recover: retry, try another source, or carry on with a note
A subagent sends every small error up to the coordinatorIt escalates too earlyRecover locally. Send up only what it couldn't fix, with the partial results
The search worked and there really are no resultsNothing failedReturn a successful empty result. It's not an error
Rule: never hide a failure, and never let one failure end the run. Send the coordinator enough detail to choose what to do next, and keep “couldn't search” separate from “searched and found nothing”.

Guide 5.3: “Structured error context (failure type, attempted query, partial results, alternatives) enables coordinator recovery. Distinguish access failures (retry) from valid empty results. Generic statuses hide context; silently suppressing errors or terminating on single failure are both anti-patterns.” Guide 2.2: “local recovery in subagents, propagate only unresolved errors with partial results.”

Example. The search subagent times out after finding 3 of an expected 10 sources. What should it return?
  1. The 3 sources, marked as a full success
  2. An exception that stops the workflow
  3. An error saying it timed out, the query it ran, the 3 sources found, and other sources worth trying
  4. Retry quietly, then return “search unavailable” if it still fails
C. The coordinator can keep the 3 and decide how to find the rest. A hides that 7 are missing. B ends the run over a recoverable problem. D works, but the coordinator learns nothing it can act on.

Mind map 6: Sources disagree or citations are lost

What you seeWhere the fault isFix
The final report has claims but no sourcesSummarizing dropped the link between claim and sourceKeep a claim-to-source map through every step
Two sources give different figuresThe agent picked one and dropped the otherReport both figures, each with its source
The figures differ because one is from 2022 and one from 2025It's a change over time, not a contradictionRequire publication dates so the difference is explained
Financial figures are buried in paragraphsEvery kind of content is written as proseShow financial data as tables and news as prose
Rule: every claim keeps its source all the way to the report. When sources disagree, show both with their sources. Check the dates before calling it a contradiction.

Guide 5.6: “Preserve structured claim-source mappings through synthesis; annotate conflicting statistics with source attribution (don't arbitrarily pick one). Require publication/collection dates so temporal differences aren't misread as contradictions; render content types appropriately (financial as tables, news as prose).”

Example. One government report says 12% of households own an EV. An industry survey says 19%. How should the research report handle it?
  1. Use the government figure, since it's more official
  2. Average them to 15.5%
  3. Show both figures, each with its source and date, and note that they differ
  4. Leave the figure out because the sources conflict
C. The reader sees both figures and where each came from. A picks one arbitrarily. B makes up a number neither source gave. D loses useful information.

Three traps that catch people

1 · “The subagents worked correctly”

When the question says this, it's telling you the fault is not in the subagents. Look at the coordinator: how it split the work, what context it passed, and how it combined the results.

2 · Subagents share nothing automatically

There's no shared memory between coordinator and subagent. An option that assumes a subagent “already knows” earlier findings is wrong. The fix is always to put them in its prompt.

3 · Task tool: the exam's answer vs today

The Agent SDK has been renaming things, but the guide says subagents are spawned with the Task tool, and "Task" must be in allowedTools. On the exam, follow the guide. More on Exam vs 2026, item 3.

How to study this

  1. Read one mind map, then open its Trap Gym page. Change the setting and watch what happens.
  2. Do the D1 practice questions and D5 practice questions. For each one, say which row of which map decides it.
  3. Print this page and cover the Fix column. Can you fill it in from memory?

Related: Domain 1 hub · Domain 5 hub · Extraction mind maps · Claude Code mind maps · MCP mind maps · all traps · CCA-F start here