appthin/dsh-skills-manager-plus

[English] | 中文 | Changelog

Skills & commands manager plugin for DeepSeek Harness. Adds a "技能与命令" (Skills & Commands) page to the left sidebar of the Settings screen, where you can view, enable/disable, edit, delete and add skills, and save frequent prompts as /commands that you invoke by typing /.

Screenshots

Skill list	Add skill

Edit skill	Commands	Edit command

Features

Install

Requires: DeepSeek Harness (DSH) ≥ 0.1.5-rc.1 (developed and verified on 0.1.6-alpha.1 and 0.2.0-rc.2), which itself needs Node.js ≥ 24.2.0 and Git ≥ 2.31.0.

One-click install from the plugin market

If you have https://github.com/dsh-market/dsh-market installed, open Settings → Plugin Market, search for dsh-skills-manager-plus and click install.

CLI

# Local directory (link install; changes apply immediately, handy for development)
dsh plugin --profile web add <this repo directory>

# Then restart dsh (or wait for the profile to hot-reload); the page appears under Settings → Skills & Commands

Verify

Open dsh web → Settings → Skills & Commands (the spark-icon sidebar item). If the page is missing:

1. Use a current Chrome/Edge — bundles can fail to load on older Chromium kernels;
2. Confirm it is installed into the active profile (dsh plugin --profile web ls);
3. If the browser console logs an error, report it with the error when opening an issue.

Update / uninstall

dsh plugin --profile web ls                          # list installed plugins
dsh plugin --profile web add <this repo directory>   # update (re-add overwrites)
dsh plugin --profile web remove dsh-skills-manager-plus # uninstall

Where things live

Skills are the skill filesystem itself — the same roots the DSH skill provider scans:

~/.dsh/skills                           user skills
~/.agents/skills                        user .agents skills
<project>/.dsh/skills                   project skills
<project>/.agents/skills                project .agents skills

Each skill is a <name>/SKILL.md (or a flat <name>.md) with YAML frontmatter. Enable/disable rewrites the disable-model-invocation and user-invocable keys; every other field (including unmodelled nested fields) is preserved byte-for-byte.

Commands live under:

$DSH_HOME/commands/<name>.md

One Markdown file per command, with name, description and optional argument-hint frontmatter; the body is the saved prompt. Files are watched, so add/edit/delete re-registers on the DSH command registry within about a second, without a restart.

The plugin keeps one small state file ($DSH_HOME/skills-manager-plus.json) tracking the import switches, disabled command names, and exactly which skills this page disabled in bulk — turning the .agents switch back on only restores those, never ones disabled by hand.

Safety

• Loopback only. The API sits on the same loopback web server; non-loopback sources get a 403.
• Writes are confined to the managed roots. A submitted path outside them is refused.
• State file is written atomically and validated on load; unparseable state resets to defaults.
• Commands go out as user messages. Invoking /command submits the prompt via agent.followup as an ordinary user message, so permissions and logging behave exactly as if the user typed it.

Development

node test/run.mjs          # all tests (88) across several suites

Tests use only Node's built-in node:test.

Layout

License

MIT