> ## Documentation Index
> Fetch the complete documentation index at: https://012e27a-docs-2.manicule.dev/llms.txt
> Use this file to discover all available pages before exploring further.
> For every page in one file, fetch https://012e27a-docs-2.manicule.dev/llms-full.txt (large).
> To search and read these docs as tools, connect the MCP server at https://012e27a-docs-2.manicule.dev/api/mcp.
> Every agent surface: https://012e27a-docs-2.manicule.dev/.well-known/llms-manifest.json

# Transparency Dock & Approval Gates (/desktop/transparency)

<!-- agent-signals: reading_time_min: 1 · est_tokens: 541 · updated: 2026-11-12 -->
Related: [Desktop Workspace Overview](/desktop/overview.md), [Agentic Canvas Mode](/desktop/agentic-mode.md), [Code Editor & Highlight-to-Ask AI](/desktop/code-editor.md), [Multi-File Visual Diff Viewer](/desktop/diff-viewer.md), [Integrated PTY Terminal](/desktop/terminal.md)

> How BhaiCode classifies agent actions and when it asks you before running them.

The **Transparency Dock** monitors every agent tool call and decides whether it runs, stages
for review, or waits for your approval.

## Risk tiers

```mermaid
flowchart TD
    OP["Agent requests a tool call"] --> TIER{"Classify risk"}
    TIER -->|"read-only"| AUTO["Runs automatically"]
    TIER -->|"workspace edit"| STAGE["Staged in the diff viewer"]
    TIER -->|"high risk / system"| GATE["Approval gate<br/>agent pauses"]
```

| Tier                       | Examples                                                                        | Behaviour                                                      |
| -------------------------- | ------------------------------------------------------------------------------- | -------------------------------------------------------------- |
| **Safe (read-only)**       | Reading files in the workspace, symbol search, `git status`, diffs              | Executes without interrupting you                              |
| **Workspace modification** | Creating files, editing code in the project                                     | Executes and stages in the [diff viewer](/desktop/diff-viewer) |
| **High risk (gated)**      | `npm install`, `docker run`, `rm -rf`, writes outside the workspace, `git push` | **Intercepted** — the agent waits for your choice              |

## The approval prompt

When a high-risk command is intercepted, the dock shows:

1. **The exact command** — command line, working directory, and flags.
2. **The reason** — why the agent believes it is needed.
3. **Your options**:

<Tabs>
<Tab title="Approve once">Runs this specific command only. The next similar command asks again.</Tab>

<Tab title="Always allow for session">Whitelists this command pattern until the current task finishes.</Tab>

<Tab title="Deny with feedback">Cancels the command and sends your explanation back to the agent so it can find another
approach.</Tab>
</Tabs>

<Tip title="Tune the noise">Approve repetitive safe patterns for the session rather than editing a config file — the
whitelist expires with the task.</Tip>