II.
AgentVersion overview
Reference · liveagentVersion:cursor:3-6-auto-review-issue-862
Cursor 3.6 Auto-review issue #862 patch overview
Inspect the raw attributes, linked wiki pages, and inbound or outbound graph edges for agentVersion:cursor:3-6-auto-review-issue-862.
Attributes
agentId
cursor
agentProductId
displayName
Cursor 3.6 Auto-review issue #862 patch
runtimeFamily
ide-agent
releaseChannel
stable
versionRange
3.6
currentVersion
3.6
releasedAt
2026-05-29
releaseNotesUrl
cliCommand
cursor-agent
summary
Cursor 3.6 Auto-review run mode policy patch for Shell, MCP, and Fetch tool calls.
assimilationNotes
- Issue #862 tracks Cursor's 2026-05-29 official changelog entry for Cursor 3.6 Auto-review Run Mode.
- Auto-review is intended to let Cursor work longer with fewer approval prompts and safer execution.
- Auto-review applies to Shell, MCP, and Fetch tool calls.
- Allowlisted calls run immediately.
- Calls that can be sandboxed run in the sandbox.
- Other agent actions route to a classifier subagent.
- The classifier subagent can allow the call, try a different approach, or ask for approval.
- Run mode is configured at Settings > Cursor Settings > Agents > Run Mode.
- The classifier agent can be steered with custom instructions.
- No public installer or package-name migration was identified for this dated Cursor changelog marker.
releaseHighlights
- Auto-review Run Mode.
- Shell, MCP, and Fetch tool-call policy scope.
- Allowlist immediate execution, sandbox fallback, and classifier-subagent decision routing.
- Classifier custom instructions through Cursor settings.
releaseTracking
issue
862
releaseMarker
3.6
changelogDate
2026-05-29
sourceKind
official-changelog
publicCliSemverIdentified
false
installerChangeIdentified
false
sharedRollupEditAvoided
true
autoReviewRunMode
runMode
auto-review
settingsPath
Settings > Cursor Settings > Agents > Run Mode
toolScopes
- Shell
- MCP
- Fetch
decisionFlow
allowlisted
run-immediately
sandboxable
run-in-sandbox
other
classifier-subagent
classifierOutcomes
- allow-call
- try-different-approach
- ask-for-approval
classifierInstructions
customInstructionsSupported
true
notes
Cursor documents that the classifier agent can be steered with custom instructions.
adapterImpact
approvalModeMapping
Auto-review is distinct from simple prompt/yolo modes because tool calls can be allowlisted, sandboxed, classifier-approved, redirected to a different approach, or escalated for approval.
toolPolicyScope
Shell, MCP, and Fetch tool calls should be modeled as covered by the Auto-review run mode.
sandboxPolicy
Sandboxable calls may execute in a sandbox before requiring approval.
Outgoing edges
sourced_from1
- evidence:cursor-3-6-auto-review-issue-862·EvidenceSourceCursor 3.6 Auto-review changelog for issue #862
version_of1
- agent:cursor·AgentProductCursor
Incoming edges
references1
- evidence:cursor-3-6-auto-review-issue-862·EvidenceSourceCursor 3.6 Auto-review changelog for issue #862