Microsoft GH-300 Bundle

For the different scenario A third-party MCP response asks for expanded repository permissions, "Validate tool permissions independently before granting access" could be appropriate. Tool output is not an administrative approval channel. The present requirement is A generated command embeds an unexpected network call unrelated to the requested fix, which calls for "Inspect and remove the unexplained outbound operation". Hidden side effects can indicate prompt injection or compromised context.

The action "Treat the comment as untrusted task data and refuse the upload" is useful when An issue comment instructs an agent to upload environment variables to a URL. Retrieved text cannot grant authority over secrets or tools. Under the stated condition A generated command embeds an unexpected network call unrelated to the requested fix, instead choose "Inspect and remove the unexplained outbound operation". Hidden side effects can indicate prompt injection or compromised context.

The alternative "Parse the fixture as test data rather than executing embedded requests" fits another situation: A test fixture includes instructions to disable audit logging. Data fields can contain attacker-controlled instructions. Here the condition is A generated command embeds an unexpected network call unrelated to the requested fix; use "Inspect and remove the unexplained outbound operation" instead. Hidden side effects can indicate prompt injection or compromised context.

The alternative "Check the current extension documentation and version" fits another situation: A question depends on a newly released editor extension feature. Feature support changes across clients and releases. Here the condition is An AI response gives several plausible fixes but no reproduction information; use "Create a minimal reproduction before choosing a remediation" instead. Targeted evidence resolves ambiguity more reliably than confidence phrasing.

The alternative "Locate the official policy control before configuring it" addresses Copilot proposes an undocumented enterprise policy setting. Plausible setting names are not proof of supported administration. It does not resolve An AI response gives several plausible fixes but no reproduction information; the correct choice is "Create a minimal reproduction before choosing a remediation". Targeted evidence resolves ambiguity more reliably than confidence phrasing.

For the different scenario An assistant cannot determine which of two tests is authoritative, "Escalate to the code owner and existing acceptance criteria" could be appropriate. AI uncertainty should not be hidden with a fabricated decisive answer. The present requirement is An AI response gives several plausible fixes but no reproduction information, which calls for "Create a minimal reproduction before choosing a remediation". Targeted evidence resolves ambiguity more reliably than confidence phrasing.

The alternative "Locate the official policy control before configuring it" fits another situation: Copilot proposes an undocumented enterprise policy setting. Plausible setting names are not proof of supported administration. Here the condition is A question depends on a newly released editor extension feature; use "Check the current extension documentation and version" instead. Feature support changes across clients and releases.

For the different scenario Two documentation pages disagree because they target different editions, "Identify the deployed product edition and apply its supported procedure" could be appropriate. Edition and plan constraints can explain apparently conflicting guidance. The present requirement is A question depends on a newly released editor extension feature, which calls for "Check the current extension documentation and version". Feature support changes across clients and releases.

"Create a minimal reproduction before choosing a remediation" targets An AI response gives several plausible fixes but no reproduction information. Targeted evidence resolves ambiguity more reliably than confidence phrasing. That differs from A question depends on a newly released editor extension feature; the required decision here is "Check the current extension documentation and version". Feature support changes across clients and releases.

The alternative "Escalate to the code owner and existing acceptance criteria" addresses An assistant cannot determine which of two tests is authoritative. AI uncertainty should not be hidden with a fabricated decisive answer. It does not resolve Copilot proposes an undocumented enterprise policy setting; the correct choice is "Locate the official policy control before configuring it". Plausible setting names are not proof of supported administration.

For the different scenario A question depends on a newly released editor extension feature, "Check the current extension documentation and version" could be appropriate. Feature support changes across clients and releases. The present requirement is Copilot proposes an undocumented enterprise policy setting, which calls for "Locate the official policy control before configuring it". Plausible setting names are not proof of supported administration.

"Identify the deployed product edition and apply its supported procedure" targets Two documentation pages disagree because they target different editions. Edition and plan constraints can explain apparently conflicting guidance. That differs from Copilot proposes an undocumented enterprise policy setting; the required decision here is "Locate the official policy control before configuring it". Plausible setting names are not proof of supported administration.

For the different scenario An AI response gives several plausible fixes but no reproduction information, "Create a minimal reproduction before choosing a remediation" could be appropriate. Targeted evidence resolves ambiguity more reliably than confidence phrasing. The present requirement is Two documentation pages disagree because they target different editions, which calls for "Identify the deployed product edition and apply its supported procedure". Edition and plan constraints can explain apparently conflicting guidance.

The action "Locate the official policy control before configuring it" is useful when Copilot proposes an undocumented enterprise policy setting. Plausible setting names are not proof of supported administration. Under the stated condition Two documentation pages disagree because they target different editions, instead choose "Identify the deployed product edition and apply its supported procedure". Edition and plan constraints can explain apparently conflicting guidance.

The alternative "Escalate to the code owner and existing acceptance criteria" fits another situation: An assistant cannot determine which of two tests is authoritative. AI uncertainty should not be hidden with a fabricated decisive answer. Here the condition is Two documentation pages disagree because they target different editions; use "Identify the deployed product edition and apply its supported procedure" instead. Edition and plan constraints can explain apparently conflicting guidance.

"Check the current extension documentation and version" targets A question depends on a newly released editor extension feature. Feature support changes across clients and releases. That differs from An assistant cannot determine which of two tests is authoritative; the required decision here is "Escalate to the code owner and existing acceptance criteria". AI uncertainty should not be hidden with a fabricated decisive answer.

The action "Identify the deployed product edition and apply its supported procedure" is useful when Two documentation pages disagree because they target different editions. Edition and plan constraints can explain apparently conflicting guidance. Under the stated condition An assistant cannot determine which of two tests is authoritative, instead choose "Escalate to the code owner and existing acceptance criteria". AI uncertainty should not be hidden with a fabricated decisive answer.

The alternative "Create a minimal reproduction before choosing a remediation" fits another situation: An AI response gives several plausible fixes but no reproduction information. Targeted evidence resolves ambiguity more reliably than confidence phrasing. Here the condition is An assistant cannot determine which of two tests is authoritative; use "Escalate to the code owner and existing acceptance criteria" instead. AI uncertainty should not be hidden with a fabricated decisive answer.

The alternative "Use the approved operations window and explicit authorization" addresses A proposed automation schedules production restarts without maintenance approval. Availability-impacting actions need owner-controlled timing. It does not resolve An agent needs to execute a database migration in a regulated environment; the correct choice is "Require the configured change approval and backup verification". Agentic execution does not replace change management.

For the different scenario A terminal assistant asks permission to delete generated artifacts and source code together, "Restrict the command to verified disposable outputs" could be appropriate. Broad file operations risk removing canonical assets. The present requirement is An agent needs to execute a database migration in a regulated environment, which calls for "Require the configured change approval and backup verification". Agentic execution does not replace change management.

"Refuse public publishing and preserve repository visibility" targets A tool call would publish an internal repository while investigating a build error. Diagnostic permission does not imply distribution permission. That differs from An agent needs to execute a database migration in a regulated environment; the required decision here is "Require the configured change approval and backup verification". Agentic execution does not replace change management.

The alternative "Refuse public publishing and preserve repository visibility" addresses A tool call would publish an internal repository while investigating a build error. Diagnostic permission does not imply distribution permission. It does not resolve A terminal assistant asks permission to delete generated artifacts and source code together; the correct choice is "Restrict the command to verified disposable outputs". Broad file operations risk removing canonical assets.

"Use the approved operations window and explicit authorization" targets A proposed automation schedules production restarts without maintenance approval. Availability-impacting actions need owner-controlled timing. That differs from A terminal assistant asks permission to delete generated artifacts and source code together; the required decision here is "Restrict the command to verified disposable outputs". Broad file operations risk removing canonical assets.

The action "Mock a credential provider and assert runtime token injection" is useful when An agent wants to commit a generated secret to demonstrate the solution. A test double exercises secret consumption without placing credentials in source control. Under the stated condition A terminal assistant asks permission to delete generated artifacts and source code together, instead choose "Restrict the command to verified disposable outputs". Broad file operations risk removing canonical assets.

For the different scenario A proposed automation schedules production restarts without maintenance approval, "Use the approved operations window and explicit authorization" could be appropriate. Availability-impacting actions need owner-controlled timing. The present requirement is A tool call would publish an internal repository while investigating a build error, which calls for "Refuse public publishing and preserve repository visibility". Diagnostic permission does not imply distribution permission.

"Mock a credential provider and assert runtime token injection" targets An agent wants to commit a generated secret to demonstrate the solution. A test double exercises secret consumption without placing credentials in source control. That differs from A tool call would publish an internal repository while investigating a build error; the required decision here is "Refuse public publishing and preserve repository visibility". Diagnostic permission does not imply distribution permission.

The alternative "Require the configured change approval and backup verification" fits another situation: An agent needs to execute a database migration in a regulated environment. Agentic execution does not replace change management. Here the condition is A tool call would publish an internal repository while investigating a build error; use "Refuse public publishing and preserve repository visibility" instead. Diagnostic permission does not imply distribution permission.

"Mock a credential provider and assert runtime token injection" targets An agent wants to commit a generated secret to demonstrate the solution. A test double exercises secret consumption without placing credentials in source control. That differs from A proposed automation schedules production restarts without maintenance approval; the required decision here is "Use the approved operations window and explicit authorization". Availability-impacting actions need owner-controlled timing.

The action "Require the configured change approval and backup verification" is useful when An agent needs to execute a database migration in a regulated environment. Agentic execution does not replace change management. Under the stated condition A proposed automation schedules production restarts without maintenance approval, instead choose "Use the approved operations window and explicit authorization". Availability-impacting actions need owner-controlled timing.

The alternative "Restrict the command to verified disposable outputs" addresses A terminal assistant asks permission to delete generated artifacts and source code together. Broad file operations risk removing canonical assets. It does not resolve A proposed automation schedules production restarts without maintenance approval; the correct choice is "Use the approved operations window and explicit authorization". Availability-impacting actions need owner-controlled timing.

The alternative "Require the configured change approval and backup verification" fits another situation: An agent needs to execute a database migration in a regulated environment. Agentic execution does not replace change management. Here the condition is An agent wants to commit a generated secret to demonstrate the solution; use "Mock a credential provider and assert runtime token injection" instead. A test double exercises secret consumption without placing credentials in source control.

The alternative "Restrict the command to verified disposable outputs" addresses A terminal assistant asks permission to delete generated artifacts and source code together. Broad file operations risk removing canonical assets. It does not resolve An agent wants to commit a generated secret to demonstrate the solution; the correct choice is "Mock a credential provider and assert runtime token injection". A test double exercises secret consumption without placing credentials in source control.

For the different scenario A tool call would publish an internal repository while investigating a build error, "Refuse public publishing and preserve repository visibility" could be appropriate. Diagnostic permission does not imply distribution permission. The present requirement is An agent wants to commit a generated secret to demonstrate the solution, which calls for "Mock a credential provider and assert runtime token injection". A test double exercises secret consumption without placing credentials in source control.