dsh-auto-approve: A DSH plugin that adds an auto-approval permission tier for DeepSeek Harness

jiao-xxx/dsh-auto-approve

Workflow & AutomationVerified
Listed on 2026-08-20
Page last updated 2026-08-20

Adds an Auto permission preset to DeepSeek Harness, auto-approving routine sandbox upgrades while routing dangerous operations to human approval.

Insert an auto tier between workspace-write and full access in DeepSeek Harness, auto-approving routine operations while routing dangerous or uncertain ones to human approval.

install
dsh plugin --profile web add github:jiao-xxx/dsh-auto-approve
Category
Workflow & Automation
Platform
DSH-Plugin
Author
jiao-xxx
Distribution
Plugin

dsh-auto-approve Key Features

Adds Auto permission presetAuto-approves routine upgradesRoutes dangerous ops to humanKeeps full audit trail

dsh-auto-approve Repository

jiao-xxx
Publisher
jiao-xxx
Repository
jiao-xxx/dsh-auto-approve
7
Stars
1
Forks
0
Watchers
0
Open issues
LanguageJavaScript
LicenseMIT License
Website
Repo last push2026-08-19 14:16:37
Repo created2026-08-14 18:21:45
Topics
dsh-plugin
Default branchmain

dsh-auto-approve Repo Summary

dsh-auto-approve is a permission management plugin (DSH plugin) for DeepSeek Harness, maintained by Jiao-XXX under the MIT license, last updated in 2026-08. It inserts an auto permission tier between the native read-only, workspace-write, and danger-full-access levels, addressing the pain point of frequent manual confirmations during routine sandbox upgrades: more convenient than workspace-write, safer than full access. Its core capability is letting a classification model auto-approve routine operations (like dependency installation or ordinary git push), while dangerous rules, uncertain cases, or timeouts fall back to human approval, keeping a full audit trail.

What core features does the dsh-auto-approve DSH plugin support?

The plugin includes a built-in dangerous command list covering destructive rm -rf targets, device writes, force pushes, download-then-execute, destructive SQL, host shutdown, recursive chmod 777, fork bombs, Terraform/Pulumi destroys, and obfuscated forms combining rm, dd, mkfs, chmod, or chown with $(), backticks, or <(). LLMs cannot override a matched dangerous rule. Ordinary git push to your own fork or working branch is a routine candidate, but pushes to main, master, release, production, or other shared/production branches require human review. The classifier only accepts real user messages (source.kind === "user") as authorization evidence, and messages over 2000 characters are not truncated or guessed but sent to human.

How to install dsh-auto-approve?

Install with the following command, then restart dsh web and select Auto in the Permissions dropdown:

bash
dsh plugin --profile web add github:jiao-xxx/dsh-auto-approve

You can also install from a local checkout:

bash
dsh plugin --profile web add ./dsh-auto-approve

To uninstall:

bash
dsh plugin --profile web remove dsh-auto-approve

Which DeepSeek Harness frontends and platforms does dsh-auto-approve support?

The plugin only depends on DSH's approval/request waterfall and permissionPresets service, independent of frontend type. Web (dsh web) fully supports the approval dialog, Auto icon compatibility layer, and /permission switching; TUI (ccch1mneyyy/dsh-TUI) supports auto-approval but requires setting permission.defaultPreset: auto in settings to enter Auto mode; Desktop (e.g., xiincs/deepseek-harness-desktop) matches the Web experience.

What are the configuration options for dsh-auto-approve?

Key options include presetName (default auto, the permission tier name) and provider (default null, using the default model provider configured in Settings → Models, compatible with any API). The plugin is pure ESM JavaScript with no build scripts, so installing from Git does not require pnpm build authorization.

View more DSH plugins at dsh-plugin.org

DSH-Plugin FAQ