GH-600 · Domain 1: Prepare agent architecture and SDLC processes · 18% of the exam
Task 1.3: Configure observability and control for autonomous agents
Setting how much autonomy an agent gets, making it leave inspectable artifacts in normal development tools, and letting people step in without slowing delivery.
Study it
Autonomy settings, session logs, commits and pull requests as inspectable artifacts
Steering, comments and approvals that keep people in control without slowing delivery
Lesson coming
Sample questions
Try each one before opening the answer. Every option is explained, with the GitHub documentation page that proves it.
Question 1 · choose 1
A tech lead reviewing a pull request that Copilot cloud agent opened wants a quick answer to "which checks did you run, and why did you change this file?" without reading the whole session log. What is the documented way to get it?
ARequest a review from Copilot code review, which reports in its comments what the agent validated
BAsk Copilot Chat on GitHub.com about the pull request, which answers from the session logs
CSearch the enterprise audit log, which records each command the agent ran
DOpen Copilot Memory, which keeps a summary of every finished session
Show the answer and why
ARequest a review from Copilot code review, which reports in its comments what the agent validated
Incorrect
Code review gives feedback on the changes; it doesn't report what the agent ran during its session.
BAsk Copilot Chat on GitHub.com about the pull request, which answers from the session logs
Correct
Copilot Chat can answer what changed, what was validated and why by drawing on the agent session logs.
CSearch the enterprise audit log, which records each command the agent ran
Incorrect
Audit log events record actions such as creating a pull request, not the agent's commands and reasoning.
DOpen Copilot Memory, which keeps a summary of every finished session
Incorrect
Copilot Memory stores repository facts and user preferences, not session summaries.
When a session is complete, or when you view a pull request Copilot created, you can ask Copilot Chat on GitHub.com what changed, what was validated and why; it answers by drawing on the session logs. The session log itself shows Copilot's reasoning and the tools it used.
An engineering manager is budgeting a pilot in which developers will redirect running agent sessions from the Agents tab, using both Copilot cloud agent and the Claude third-party agent. Which statement belongs in the pilot guide?
ASteering is free, because it continues a session that is already paid for
BSteering pauses the agent until a reviewer approves the redirected plan
CEach steering message costs AI credits, and partner agents such as Claude can't be steered
DSteering works for every agent listed in the Agents tab, including Claude and Codex sessions
Show the answer and why
ASteering is free, because it continues a session that is already paid for
Incorrect
Each steering message consumes AI credits.
BSteering pauses the agent until a reviewer approves the redirected plan
Incorrect
Steering doesn't add an approval step; Copilot acts on the message while the session keeps running.
CEach steering message costs AI credits, and partner agents such as Claude can't be steered
Correct
Steering consumes AI credits per message and isn't available for third-party coding agents.
DSteering works for every agent listed in the Agents tab, including Claude and Codex sessions
Incorrect
Steering isn't available for third-party coding agents.
From the Agents tab or the agents page you can give steering input to a running Copilot cloud agent session without stopping it, for example to point it at an existing utility class. Each steering message consumes AI credits, and steering is not available for third-party coding agents.
An organization already runs its own static analysis product on every pull request. A repository administrator finds that Copilot cloud agent's built-in CodeQL scan and second-opinion review duplicate that work and slow each session down. What should the administrator change?
ASwitch off the overlapping tools in the repository's cloud agent settings
BDisable the agent firewall so the scans can reach the external product
CTurn off Require approval for workflow runs so the existing checks start sooner
DRemove the GitHub Advanced Security license from the repository
Show the answer and why
ASwitch off the overlapping tools in the repository's cloud agent settings
Correct
Repository administrators can disable the built-in code quality and security validation tools to work faster or avoid conflicts with other products.
BDisable the agent firewall so the scans can reach the external product
Incorrect
The firewall limits internet access; it doesn't control the built-in validation tools.
CTurn off Require approval for workflow runs so the existing checks start sooner
Incorrect
That setting controls whether Actions workflows run on Copilot's pushes; it doesn't touch the validation tools and adds risk.
DRemove the GitHub Advanced Security license from the repository
Incorrect
Licensing isn't how the validation tools are switched off; they have their own toggles in the repository's cloud agent settings.
By default, Copilot cloud agent checks the code it generates for security issues and gets a second opinion from Copilot code review, and tries to fix what they find before it completes the pull request. A repository administrator can disable one or more of these validation tools in the repository's Copilot cloud agent settings, for example to work faster or to avoid conflicts with other code quality or security products.
A developer wants to clear a finished Copilot cloud agent session out of their sessions list. What can they do with it?
AArchive it, because cloud agent sessions can't be deleted
BDelete it permanently from the session menu
CRewind it with /undo to remove its history
DClose its pull request, which removes the session from every list
Show the answer and why
AArchive it, because cloud agent sessions can't be deleted
Correct
Cloud agent sessions can be archived but not deleted.
BDelete it permanently from the session menu
Incorrect
Only local sessions can be deleted.
CRewind it with /undo to remove its history
Incorrect
Rewinding changes a session's history; it doesn't remove the session from your list.
DClose its pull request, which removes the session from every list
Incorrect
Closing a pull request doesn't remove its session.
You can archive stopped Copilot cloud agent sessions to remove them from your sessions list. They can be archived but not deleted; only local sessions from Copilot CLI, VS Code, JetBrains IDEs or the GitHub Copilot app can be deleted.
A developer shares a local Copilot CLI session from the repository's Agents tab. What can collaborators do with that session?
ASteer it with new prompts while it runs
BResume it on their own machines as their session
CEdit its file changes before they are pushed
DView its prompts, responses and file changes
Show the answer and why
ASteer it with new prompts while it runs
Incorrect
Recipients cannot steer or modify a shared session.
BResume it on their own machines as their session
Incorrect
Sharing gives view-only access, not a copy to resume.
CEdit its file changes before they are pushed
Incorrect
Recipients cannot modify a shared session.
DView its prompts, responses and file changes
Correct
Recipients can view the session's prompts, responses and file changes.
Local sessions from Copilot CLI, VS Code, JetBrains IDEs or the GitHub Copilot app are unshared by default. You can share one from the repository's Agents tab, giving repository collaborators view-only access to its prompts, responses and file changes; they cannot steer or modify it.
A reviewer wants to know which custom agent produced a pull request that Copilot opened. Where does Copilot record that?
AOnly in the enterprise audit log
BIn each commit's author field
CIn the pull request description
DIn a label that Copilot adds to the pull request
Show the answer and why
AOnly in the enterprise audit log
Incorrect
The pull request description itself records the custom agent.
BIn each commit's author field
Incorrect
The commit author field doesn't name a custom agent; the description does.
CIn the pull request description
Correct
When Copilot opens pull requests, it notes which custom agent was used in the description.
DIn a label that Copilot adds to the pull request
Incorrect
No label is documented for this; the description records it.
You can use custom agents wherever Copilot cloud agent is available, for example by picking one when you assign an issue. When Copilot opens a pull request, it notes which custom agent was used to complete the work in the pull request description.
A developer turns on remote control so they can follow and steer a Copilot CLI session from GitHub Mobile while away from their desk. What must remain true for this to work?
AThe session must first be moved to a cloud sandbox
BThe CLI must keep running on a machine that stays online
CThe working directory must be a repository hosted on GitHub.com
DThe session must have been started with --prompt
Show the answer and why
AThe session must first be moved to a cloud sandbox
Incorrect
Remote control works with a session running locally; no cloud sandbox is needed.
BThe CLI must keep running on a machine that stays online
Correct
The session continues on the original machine, which must be online with the CLI running.
CThe working directory must be a repository hosted on GitHub.com
Incorrect
Remote control doesn't require a GitHub-hosted repository.
DThe session must have been started with --prompt
Incorrect
Remote control is only for interactive sessions, not programmatic --prompt runs.
Remote control lets you monitor and steer a Copilot CLI session from GitHub.com or GitHub Mobile. The session keeps running on your machine, which must stay online; it is available only for interactive sessions, and for organization seats the "Store local sessions in the Cloud" policy must be set to View and control.