Claude Code Auto Mode: Denied Commands Now Visible and Retryable in `/permissions`
Claude CodeClaude Code v2.1.89 improves the transparency of Auto mode by surfacing denied tool calls directly in the /permissions UI. Users can now review which commands were blocked and retry them after adjusting their permission settings, eliminating the need to restart tasks to recover from a denied operation.
Sources & Mentions
3 external resources covering this update
Auto Mode Gets a Transparency Upgrade
Claude Code Auto mode allows the AI to execute commands and tool calls autonomously, but until now, any command denied by permission rules would simply fail silently — leaving users uncertain about what was blocked and why. Version 2.1.89 changes this by making denied tool calls visible within the /permissions interface.
Denied Commands Now Surface in /permissions
When Claude Code operates in Auto mode and encounters a tool call blocked by the user's permission settings, that denial is now logged and surfaced in the /permissions view. Users can open this panel during or after a session to see exactly which commands were refused and understand what triggered the restriction.
This transparency is particularly valuable in long-running agentic sessions where the AI may have attempted several operations before the user checks in. Instead of discovering a partially completed task with no indication of what went wrong, users can now audit the full list of denied actions in one place.
Retry Without Restarting
The more impactful half of this change is the retry capability. Once a denied command is visible in /permissions, users can adjust their permission rules and immediately retry the blocked operation — without abandoning the current session or re-running the task from scratch.
Previously, recovering from a permission denial often meant restarting the entire workflow. This friction discouraged users from running Claude Code in tighter permission modes, since a single denial could derail a session. The retry flow removes that penalty, making it practical to start with more restrictive permissions and loosen them incrementally as trust is established.
Why This Matters for Agentic Workflows
As Claude Code is increasingly used for multi-step autonomous tasks, the ability to inspect and correct mid-session failures becomes critical. This update moves the permissions system from a passive blocklist into an active diagnostic and recovery tool — a meaningful step toward more controllable agentic AI.