Add "Blocklist Mode" & Workspace-Aware Agent Safety Settings

Feature request for product/service

Chat

Describe the request

Problem

Cursor has great Auto-Run settings (File-Deletion Protection, Dotfile Protection, etc.), but there are critical gaps:

1. Allowlist is Too Restrictive

  • Current “Use Allowlist” mode requires whitelisting EVERY command (ls, pwd, npm, git status…)
  • This breaks normal workflows and is tedious to maintain
  • What we need: “Use Blocklist” - run everything EXCEPT specific dangerous operations

2. Missing Git-Specific Protection Toggles

  • File protections exist (deletion, dotfile, external-file) :white_check_mark:
  • Git protections missing (commit, push, branch operations) :cross_mark:
  • Git operations can happen automatically without safeguards

3. Settings Are Global, Not Workspace-Aware

  • Auto-Run settings apply to ALL workspaces
  • Production repos need stricter rules than personal projects
  • No way to have different safety levels per workspace

4. Inconsistent Agent Behavior Across Sessions

  • Different chat sessions show different “aggression levels”
  • One agent auto-commits, another always asks
  • No persistent workspace-specific agent behavior

Proposed Solution

1. Add “Blocklist Mode” to Auto-Run Settings

Add a third option to the Auto-Run dropdown:

  • :white_check_mark: Run everything (current)
  • :white_check_mark: Use Allowlist (current - too restrictive)
  • :plus: Use Blocklist (NEW - run everything EXCEPT dangerous operations)

Example Blocklist:

Command Blocklist: rm -rf, sudo, dd, ssh production*

Real-World Use Case:

  • Developer wants normal commands to work (ls, npm install, git status)
  • But block dangerous ones (git commit, ssh *, rm -rf, sudo)
  • Maintains productivity with critical guardrails

2. Add Git Protection Toggles

Extend existing protection toggles with git-specific options:

Current:

  • :check_box_with_check: File-Deletion Protection
  • :check_box_with_check: Dotfile Protection
  • :check_box_with_check: External-File Protection

Proposed:

  • :plus: Git Commit Protection - Require confirmation before committing
  • :plus: Git Push Protection - Confirm before pushing to remote
  • :plus: Git Branch Protection - Confirm branch creation/deletion
  • :plus: SSH Connection Protection - Require approval for SSH commands

3. Make Auto-Run Settings Workspace-Aware

Current: Auto-Run settings are global across all workspaces
Proposed: Save settings per-workspace (.cursor/auto-run-config.json)

Why This Matters:

Personal Project:
→ Auto-Run: Use Blocklist (only block rm -rf, sudo)
→ Git Protection: OFF (move fast)

Production Client Workspace:
→ Auto-Run: Use Allowlist (very conservative)
→ Git Protection: ALL ON (never auto-commit/push)
→ SSH Protection: ON (never connect without approval)

Implementation:

  • Add “Use Workspace Settings” toggle in Auto-Run panel
  • Settings saved to .cursor/auto-run-config.json in workspace
  • Visual indicator: “:locked: Workspace Safety Rules Active”
  • Team can commit workspace config to git for consistency

Benefits

  • Blocklist Mode: Maintains productivity while adding guardrails (best of both worlds)
  • Git Protection: Prevents accidental commits/pushes (extends existing protection pattern)
  • Workspace-Aware: Different projects, different risk profiles, different settings
  • Cross-Session Consistency: Agent behavior stays consistent within workspace
  • Enterprise-Ready: Teams can enforce safety policies via committed workspace configs

Real-World Context

We’re a web development agency using Cursor extensively across our team. We currently use “Run Everything” (too risky) because “Use Allowlist” is too tedious. A blocklist mode would let us maintain workflow speed while adding critical safety guardrails.

I’ll be presenting about AI-assisted workflows (featuring Cursor) at the EE Conference in London (Oct 15, 2025) and would love to showcase these safety features to the developer audience.


Additional References:

  • Current Auto-Run Settings: Good foundation to build on
  • Repository Blocklist (Enterprise): Proves Cursor understands blocklist UX
  • File Protection Toggles: Excellent pattern to extend to git/SSH operations
1 Like

I fully agree with this; not being able to block/allow commands per workspace is a huge blocker for me utilising AI features fully.

In most of my projects I have 100% test coverage requirements, only allowing merges to main etc setup, so they could be using AI features safely even now.

However, in some of the projects I can’t for reasons out of my control prevent pushing straight to main etc, which leads to not being able to utilise AI fully in other workspaces.

I also agree that there’s a need for having more granual control, for example allowing `git commit` but not `git push` etc. This also would benefit issues raised in other topics, for example when running Python scripts. Now Cursor asks for permission to run Python, which is terrible for security. Instead allowing only `python utils.py` or `python -m unittest` etc would be a whole lot safer, even though Cursor of course could still modify those tools as it wishes.

1 Like