Auto-Approving Actions
Every auto-approve toggle and its recommended setting
Nexgile Code is approval-first: every action that touches your files, your terminal, or an external service stops and asks before it runs. That default is the right one for a new project or an unfamiliar codebase — but once you trust a category of action, clicking Approve forty times per task is friction without safety. The auto-approve settings let you remove prompts one category at a time, while keeping hard limits and command guardrails in force.
This page covers the master switch, the seven category toggles, their modifiers, and a recommended starting profile.
Where the controls live
Section titled “Where the controls live”- Settings → Auto-Approve — the full panel: master switch, category toggles, modifiers, command lists, and request/cost limits.
- The Auto-Approve control in the chat panel — a compact dropdown with the same category switches, plus Select All / Select None. Turning on any category here also turns on the master switch.
- Keyboard:
Ctrl+Alt+A(Cmd+Alt+Aon macOS) — runs Toggle Auto-Approve, flipping the master switch without leaving the keyboard. Use it to drop into fully supervised mode before a risky step, then flip back.
The master switch
Section titled “The master switch”Auto-Approve Enabled (autoApprovalEnabled) gates everything. When it is off, every category toggle is ignored and Nexgile Code asks for every action. When it is on, only the categories you have individually enabled skip their prompts — everything else still asks.
The seven toggles
Section titled “The seven toggles”| Toggle (label) | Setting key | What it unlocks when on |
|---|---|---|
| Read | alwaysAllowReadOnly |
Reading files and listing directory contents inside the workspace, without a prompt |
| Write | alwaysAllowWrite |
Creating and editing files inside the workspace, without a prompt |
| Execute | alwaysAllowExecute |
Running terminal commands that match your allowlist (see below) |
| MCP | alwaysAllowMcp |
Running MCP tools — but only tools you have also marked “Always allow” on their server in the MCP Servers view |
| Mode | alwaysAllowModeSwitch |
Switching between modes (for example Code → Debug) without a prompt |
| Subtasks | alwaysAllowSubtasks |
Creating and completing subtasks without a prompt |
| Question | alwaysAllowFollowupQuestions |
Auto-answering the agent’s follow-up questions with the first suggested answer after a timeout |
Two important interactions:
- Execute is not “run anything.” With Execute on, each command is still checked against your Allowed Auto-Execute Commands and Denied Commands lists. Commands that match neither list fall back to a prompt. See Command Guardrails & Limits.
- MCP needs two keys to turn. The MCP toggle alone approves resource access; for tool calls, the specific tool must also have its “Always allow” checkbox set on that server. A newly added MCP server never runs unattended by accident.
Modifiers under each toggle
Section titled “Modifiers under each toggle”Enabling Read, Write, or Question reveals additional controls:
| Modifier | Setting key | Applies to | Default |
|---|---|---|---|
| Include files outside workspace | alwaysAllowReadOnlyOutsideWorkspace |
Read | Off |
| Include files outside workspace | alwaysAllowWriteOutsideWorkspace |
Write | Off |
| Include protected files | alwaysAllowWriteProtected |
Write | Off |
| Delay after writes | writeDelayMs |
Write | 1000 ms |
| Time to wait before auto-selecting the first answer | followupAutoApproveTimeoutMs |
Question | 60 s (slider 1–300 s) |
- Outside-workspace modifiers. By default, auto-approval stops at the workspace boundary — a read or write that reaches outside the open folder always prompts, even with the category toggle on. Only enable these if you genuinely need the agent working across folders.
- Protected files. Files like
.nexgileignore,.nexgilerules*,.nexgile/**,.nexgilemodes,.vscode/**, andAGENTS.mdare configuration that shapes the agent’s own behavior. Writes to them always prompt unless you enable this modifier. Leave it off. See .nexgileignore & Protected Files. - Write delay. After each auto-approved write, Nexgile Code waits (1 second by default, configurable 0–5000 ms in Settings → Context) so the editor’s diagnostics can flag new problems before the agent moves on. Keep a non-zero delay — it lets the agent catch its own compile errors.
- Follow-up timeout. When the agent asks a question and offers suggested answers, the Question toggle auto-selects the first suggestion after the timeout (60 seconds by default). You can always answer manually before the timer expires.
Always auto-approved
Section titled “Always auto-approved”Two actions never prompt, regardless of your settings:
- Todo list updates — the agent maintaining its own work plan is bookkeeping, not a side effect.
- Skill loads — the
skilltool only loads instruction packages you (or your project) explicitly installed, so loading one is treated like reading your own configuration. See Skills.
Request and cost caps still apply
Section titled “Request and cost caps still apply”Auto-approval never overrides the hard limits set in the same panel: Max Count (allowedMaxRequests) and Max Cost (allowedMaxCost). When a task reaches either cap, it pauses with an “Auto-Approved Limit Reached” notice and waits for you to click Reset and Continue. Details in Command Guardrails & Limits.
Subtasks inherit the parent’s autonomy
Section titled “Subtasks inherit the parent’s autonomy”When a task that is running with the master switch on spawns a child task (for example, Orchestrator delegating to Code, or a Workflow Mode chain), the child inherits full autonomy for the duration of its run — all category toggles are treated as on for the child’s own approval checks. Nothing is changed in your saved settings; the promotion applies only while the child runs on the parent’s behalf.
This is what makes “set it and walk away” delegation work: you opted into autonomy at the top level, and the children the parent creates carry that opt-in. Top-level tasks you start yourself always use your exact settings. Command allow/deny lists and the request/cost caps still apply to children. See Subtasks & Todo Lists and Workflow Mode.
A “start conservative” profile
Section titled “A “start conservative” profile”A sensible first configuration — fast where it is safe, prompting where it matters:
| Setting | Recommended | Why |
|---|---|---|
| Auto-Approve Enabled | On | Enables the toggles below |
| Read | On | Reading is non-destructive; this removes most prompt fatigue |
| Read outside workspace | Off | Keep the agent’s eyes inside the project |
| Write | On for trusted projects, Off while evaluating | Checkpoints make writes recoverable — but review diffs first on new codebases |
| Write outside workspace | Off | A stray write to dotfiles or another repo is the costliest mistake available |
| Write protected files | Off | Agent configuration changes deserve eyes |
| Execute | On, with a short curated allowlist | Start with read-only commands: git log, git diff, git show (the defaults) |
| MCP | Off globally; “Always allow” per vetted tool | Approve per tool, per server |
| Mode | On | Mode switches are low-risk and frequent |
| Subtasks | On | Required for smooth delegation |
| Question | Off at first | Answer the agent’s questions yourself until you trust its defaults |
| Max Cost | $5 | Circuit breaker for runaway tasks |
Tighten or loosen one toggle at a time, and re-tighten with Ctrl+Alt+A whenever a task enters territory you want to supervise.