100% Latest & Updated Microsoft AB-620 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 320
In a pharmaceutical provider, a custom API tool can delete records, although the agent only needs read access. The team selected this design: Separate privileged deletion actions from ordinary retrieval capability. Which implementation action BEST puts it into practice?
- Use the lowest-privilege connection and only approve the necessary read tools.
- Use a read-only API role and exclude destructive operations from the agent's available tools.
- Add bounded exponential backoff or queueing and respect provider retry guidance.
- Map nested provider fields into documented typed fields and explicit failure states.
Explanation
Correct Answer
B:Use a read-only API role and exclude destructive operations from the agent's available tools. The expected result is that a user request to delete data cannot be satisfied using the read-only tool.
Incorrect Answers
A:“Use the lowest-privilege connection and only approve the necessary read tools” could be appropriate when an MCP server exposes both account lookup and fund-transfer tools while the agent should only retrieve balances. The present task instead requires “Use a read-only API role and exclude destructive operations from the agent's available tools”.
C:An API responds with HTTP 429 when too many agent calls occur concurrently. That condition could justify “Add bounded exponential backoff or queueing and respect provider retry guidance”. Here, “Use a read-only API role and exclude destructive operations from the agent's available tools” targets the needed outcome.
D:“Map nested provider fields into documented typed fields and explicit failure states” could be appropriate when an agent needs to call an existing weathered internal API with a nonstandard response shape. The present task instead requires “Use a read-only API role and exclude destructive operations from the agent's available tools”.
Question 321
In a pharmaceutical provider, a custom API tool can delete records, although the agent only needs read access. After the team implements this action: Use a read-only API role and exclude destructive operations from the agent's available tools. Which evidence BEST verifies the requirement?
- Burst tests complete without an uncontrolled retry storm.
- Generated responses and transcripts contain no client secret.
- A user request to delete data cannot be satisfied using the read-only tool.
- A test conversation invokes the expected MCP tool with correctly shaped arguments.
Explanation
Correct Answer
C:A user request to delete data cannot be satisfied using the read-only tool. This result tests the effect of using a read-only API role and excluding destructive operations from the agent's available tools.
Incorrect Answers
A:An API responds with HTTP 429 when too many agent calls occur concurrently. In that setting, “Burst tests complete without an uncontrolled retry storm” is relevant; the required choice here is “A user request to delete data cannot be satisfied using the read-only tool”.
B:A REST action requires OAuth authorization, but the team proposes embedding a client secret in prompts. That condition could justify “Generated responses and transcripts contain no client secret”. Here, “A user request to delete data cannot be satisfied using the read-only tool” targets the needed outcome.
D:A Copilot Studio agent must invoke a supported order lookup exposed through a published MCP server. In that setting, “A test conversation invokes the expected MCP tool with correctly shaped arguments” is relevant; the required choice here is “A user request to delete data cannot be satisfied using the read-only tool”.
Question 322
In a helpdesk team, a user asks the agent to run a workflow that is not attached as an available tool. Which solution design decision BEST addresses this requirement?
- Treat webpage text as untrusted task data rather than new agent authority.
- Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it.
- Configure secure authentication for the API tool outside prompt content.
- Align the tool schema with the server's required arguments.
Explanation
Correct Answer
B:Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it. Implement this design by adding the supported tool, parameter schema and authorization before referencing it in instructions.
Incorrect Answers
A:A computer-use tool opens an untrusted webpage containing instructions to exfiltrate data. That condition could justify “Treat webpage text as untrusted task data rather than new agent authority”. Here, “Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it” targets the needed outcome.
C:A REST action requires OAuth authorization, but the team proposes embedding a client secret in prompts. In that setting, “Configure secure authentication for the API tool outside prompt content” is relevant; the required choice here is “Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it”.
D:An MCP service can invoke an update function but requires a field absent from the agent's tool schema. That condition could justify “Align the tool schema with the server's required arguments”. Here, “Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it” targets the needed outcome.
Question 323
In a helpdesk team, a user asks the agent to run a workflow that is not attached as an available tool. The team selected this design: Explicitly register the workflow as a tool rather than instruct the agent to hallucinate it. Which implementation action BEST puts it into practice?
- Use the lowest-privilege connection and only approve the necessary read tools.
- Use supported token renewal and return an authentication-required error when needed.
- Return a typed unavailable/error state on timeout instead of coercing to zero.
- Add the supported tool, parameter schema and authorization before referencing it in instructions.
Explanation
Correct Answer
D:Add the supported tool, parameter schema and authorization before referencing it in instructions. The expected result is that the activity trace shows a real tool call rather than an invented completion.
Incorrect Answers
A:“Use the lowest-privilege connection and only approve the necessary read tools” could be appropriate when an MCP server exposes both account lookup and fund-transfer tools while the agent should only retrieve balances. The present task instead requires “Add the supported tool, parameter schema and authorization before referencing it in instructions”.
B:An OAuth-backed tool's token expires while an agent is processing a multi-step request. In that setting, “Use supported token renewal and return an authentication-required error when needed” is relevant; the required choice here is “Add the supported tool, parameter schema and authorization before referencing it in instructions”.
C:The agent calls a stock API but occasionally treats a timeout as a zero stock count. That condition could justify “Return a typed unavailable/error state on timeout instead of coercing to zero”. Here, “Add the supported tool, parameter schema and authorization before referencing it in instructions” targets the needed outcome.
Question 324
In a helpdesk team, a user asks the agent to run a workflow that is not attached as an available tool. After the team implements this action: Add the supported tool, parameter schema and authorization before referencing it in instructions. Which evidence BEST verifies the requirement?
- A prompt requesting transfer cannot invoke a transfer operation with the read-only credentials.
- Expired credentials do not cause the agent to silently switch to a privileged maker identity.
- The activity trace shows a real tool call rather than an invented completion.
- Burst tests complete without an uncontrolled retry storm.
Explanation
Correct Answer
C:The activity trace shows a real tool call rather than an invented completion. This result tests the effect of adding the supported tool, parameter schema and authorization before referencing it in instructions.
Incorrect Answers
A:An MCP server exposes both account lookup and fund-transfer tools while the agent should only retrieve balances. In that setting, “A prompt requesting transfer cannot invoke a transfer operation with the read-only credentials” is relevant; the required choice here is “The activity trace shows a real tool call rather than an invented completion”.
B:An OAuth-backed tool's token expires while an agent is processing a multi-step request. That condition could justify “Expired credentials do not cause the agent to silently switch to a privileged maker identity”. Here, “The activity trace shows a real tool call rather than an invented completion” targets the needed outcome.
D:An API responds with HTTP 429 when too many agent calls occur concurrently. In that setting, “Burst tests complete without an uncontrolled retry storm” is relevant; the required choice here is “The activity trace shows a real tool call rather than an invented completion”.
Question 325
In a cloud administrator, a REST API uses pagination and the agent answers from only the first page of records. Which solution design decision BEST addresses this requirement?
- Handle rate limiting and avoid uncontrolled retries.
- Improve tool metadata and restrict each tool to a distinct operation.
- Distinguish unavailable data from valid numeric zero in tool output handling.
- Implement pagination logic rather than assuming the first response is complete.
Explanation
Correct Answer
D:Implement pagination logic rather than assuming the first response is complete. Implement this design by using continuation tokens until the approved result boundary is reached.
Incorrect Answers
A:An API responds with HTTP 429 when too many agent calls occur concurrently. That condition could justify “Handle rate limiting and avoid uncontrolled retries”. Here, “Implement pagination logic rather than assuming the first response is complete” targets the needed outcome.
B:“Improve tool metadata and restrict each tool to a distinct operation” could be appropriate when two API tools have near-identical descriptions, and the agent selects the wrong record-updater. The present task instead requires “Implement pagination logic rather than assuming the first response is complete”.
C:The agent calls a stock API but occasionally treats a timeout as a zero stock count. In that setting, “Distinguish unavailable data from valid numeric zero in tool output handling” is relevant; the required choice here is “Implement pagination logic rather than assuming the first response is complete”.
Question 326
In a cloud administrator, a REST API uses pagination and the agent answers from only the first page of records. The team selected this design: Implement pagination logic rather than assuming the first response is complete. Which implementation action BEST puts it into practice?
- Refresh the advertised parameter contract and validate required fields before invocation.
- Use a read-only API role and exclude destructive operations from the agent's available tools.
- Use continuation tokens until the approved result boundary is reached.
- Inspect the failed step and modify the interaction sequence, then retest it.
Explanation
Correct Answer
C:Use continuation tokens until the approved result boundary is reached. The expected result is that the reported count matches all authorized result pages, not just the first page.
Incorrect Answers
A:“Refresh the advertised parameter contract and validate required fields before invocation” could be appropriate when an MCP service can invoke an update function but requires a field absent from the agent's tool schema. The present task instead requires “Use continuation tokens until the approved result boundary is reached”.
B:A custom API tool can delete records, although the agent only needs read access. In that setting, “Use a read-only API role and exclude destructive operations from the agent's available tools” is relevant; the required choice here is “Use continuation tokens until the approved result boundary is reached”.
D:“Inspect the failed step and modify the interaction sequence, then retest it” could be appropriate when a computer-use task fails because a button moved after a UI redesign. The present task instead requires “Use continuation tokens until the approved result boundary is reached”.
Question 327
In a cloud administrator, a REST API uses pagination and the agent answers from only the first page of records. After the team implements this action: Use continuation tokens until the approved result boundary is reached. Which evidence BEST verifies the requirement?
- Requests rejected for missing fields no longer reach the MCP server.
- A test conversation invokes the expected MCP tool with correctly shaped arguments.
- The reported count matches all authorized result pages, not just the first page.
- A simulated timeout is reported as unavailable, not sold out.
Explanation
Correct Answer
C:The reported count matches all authorized result pages, not just the first page. This result tests the effect of using continuation tokens until the approved result boundary is reached.
Incorrect Answers
A:An MCP service can invoke an update function but requires a field absent from the agent's tool schema. In that setting, “Requests rejected for missing fields no longer reach the MCP server” is relevant; the required choice here is “The reported count matches all authorized result pages, not just the first page”.
B:A Copilot Studio agent must invoke a supported order lookup exposed through a published MCP server. That condition could justify “A test conversation invokes the expected MCP tool with correctly shaped arguments”. Here, “The reported count matches all authorized result pages, not just the first page” targets the needed outcome.
D:The agent calls a stock API but occasionally treats a timeout as a zero stock count. In that setting, “A simulated timeout is reported as unavailable, not sold out” is relevant; the required choice here is “The reported count matches all authorized result pages, not just the first page”.
Question 328
In a mobile-app support team, an OAuth-backed tool's token expires while an agent is processing a multi-step request. Which solution design decision BEST addresses this requirement?
- Handle authentication expiry with a secure reconnect or explicit reauthorization path.
- Distinguish unavailable data from valid numeric zero in tool output handling.
- Improve tool metadata and restrict each tool to a distinct operation.
- Align the tool schema with the server's required arguments.
Explanation
Correct Answer
A:Handle authentication expiry with a secure reconnect or explicit reauthorization path. Implement this design by using supported token renewal and return an authentication-required error when needed.
Incorrect Answers
B:“Distinguish unavailable data from valid numeric zero in tool output handling” could be appropriate when the agent calls a stock API but occasionally treats a timeout as a zero stock count. The present task instead requires “Handle authentication expiry with a secure reconnect or explicit reauthorization path”.
C:Two API tools have near-identical descriptions, and the agent selects the wrong record-updater. In that setting, “Improve tool metadata and restrict each tool to a distinct operation” is relevant; the required choice here is “Handle authentication expiry with a secure reconnect or explicit reauthorization path”.
D:An MCP service can invoke an update function but requires a field absent from the agent's tool schema. That condition could justify “Align the tool schema with the server's required arguments”. Here, “Handle authentication expiry with a secure reconnect or explicit reauthorization path” targets the needed outcome.
Question 329
In a mobile-app support team, an OAuth-backed tool's token expires while an agent is processing a multi-step request. The team selected this design: Handle authentication expiry with a secure reconnect or explicit reauthorization path. Which implementation action BEST puts it into practice?
- Constrain navigation and tool actions to the approved user task and resist page-borne instructions.
- Use supported token renewal and return an authentication-required error when needed.
- Send only required fields through an authenticated endpoint and redact unneeded attributes.
- Give each tool a clear action-oriented description, schema and permissions.
Explanation
Correct Answer
B:Use supported token renewal and return an authentication-required error when needed. The expected result is that expired credentials do not cause the agent to silently switch to a privileged maker identity.
Incorrect Answers
A:“Constrain navigation and tool actions to the approved user task and resist page-borne instructions” could be appropriate when a computer-use tool opens an untrusted webpage containing instructions to exfiltrate data. The present task instead requires “Use supported token renewal and return an authentication-required error when needed”.
C:A tool sends email addresses to an external REST endpoint but lacks purpose-limited field handling. That condition could justify “Send only required fields through an authenticated endpoint and redact unneeded attributes”. Here, “Use supported token renewal and return an authentication-required error when needed” targets the needed outcome.
D:“Give each tool a clear action-oriented description, schema and permissions” could be appropriate when two API tools have near-identical descriptions, and the agent selects the wrong record-updater. The present task instead requires “Use supported token renewal and return an authentication-required error when needed”.
Question 330
In a mobile-app support team, an OAuth-backed tool's token expires while an agent is processing a multi-step request. After the team implements this action: Use supported token renewal and return an authentication-required error when needed. Which evidence BEST verifies the requirement?
- Burst tests complete without an uncontrolled retry storm.
- Expired credentials do not cause the agent to silently switch to a privileged maker identity.
- The reported count matches all authorized result pages, not just the first page.
- High-risk tool calls are not selected under ambiguous descriptions.
Explanation
Correct Answer
B:Expired credentials do not cause the agent to silently switch to a privileged maker identity. This result tests the effect of using supported token renewal and return an authentication-required error when needed.
Incorrect Answers
A:An API responds with HTTP 429 when too many agent calls occur concurrently. In that setting, “Burst tests complete without an uncontrolled retry storm” is relevant; the required choice here is “Expired credentials do not cause the agent to silently switch to a privileged maker identity”.
C:“The reported count matches all authorized result pages, not just the first page” could be appropriate when a REST API uses pagination and the agent answers from only the first page of records. The present task instead requires “Expired credentials do not cause the agent to silently switch to a privileged maker identity”.
D:An MCP server exposes tool names but provides insufficient descriptions of side effects. In that setting, “High-risk tool calls are not selected under ambiguous descriptions” is relevant; the required choice here is “Expired credentials do not cause the agent to silently switch to a privileged maker identity”.
Designing and Building Integrated AI Agent Solutions in Copilot Studio (AB-620) Practice Tests