Preface¶
In the DeepSeek Harness (DSH) Web client, adjusting “Thinking Intensity” or reasoning levels typically involves accessing a secondary menu via the model entry point next to the input box and selecting an option level by level. To see which level is currently in use or which other levels are available, you often have to expand the menu. If you frequently switch between models and intensity levels within the same interface and prefer an entry point that functions like a native control but expands into a draggable slider, the third-party client plugin dsh-codex-power-slider (maintained by NNNOOOOMII) is designed for exactly this use case. It takes over the model entry point next to the composer, using a Codex desktop app-style card slider to read and write the actual reasoning intensity of the session. It requires zero build steps and zero Node dependencies, and is exclusively for the dsh web platform.
What is this?¶
dsh-codex-power-slider is a third-party client plugin registered on the DSH Web GUI. It replicates the interaction and visual materials of Codex. It is not affiliated with the official Codex or OpenAI, nor does it provide model services. It masks the native ModelSelect at the conversation.input.model slot with priority: -1. When collapsed, it displays “Model Name + Current Level + Dropdown Arrow”; when clicked, it expands into an intensity card with ticks, allowing access to a model list grouped by provider. Level names and selectable ranges come from adapter metadata; the slider is hidden if fewer than two levels are available. The project is open-source on GitHub (MIT), categorized as “Fun Customization” in the community directory, and currently has approximately 8 stars.
Core Features¶
Reasoning Intensity Slider and Session Writing¶
The floating card is 262px wide with a 16px radius. The center displays the current level title (14px), with a clickable model name below it. The slider track is 22px, featuring a 25px white drag handle and a tick mark for each level. Dragging or clicking the track selects an intensity; releasing the handle writes the value to the current session, and the next request will use the new level. Clicking the reset icon in the top right returns to the default level published by the model. When the slider is focused, use ↑ ↓ ← → to switch levels one by one.
Model List and Chinese Descriptions¶
Clicking the model name in the card opens a separate rounded floating list grouped by provider, with a checkmark on the current item. Hovering over an entry reveals a Chinese introduction bubble on the right (width fits the text, not exceeding 300px). Selecting an item writes the session model and intensity projection, sharing the same modelSelection projection as the native model menu.
Ultra Level Visual Feedback¶
When a model has at least three intensity levels and is stopped at the highest level (e.g., ultra), the Ultra skin is activated: a 1.8s two-stage purple scan, 24 stardust particles revealed during the scan, 8 dark blue-purple ink splatters emitted from the drag handle (disappearing within ~500ms), and multi-layered continuous flowing aurora backlighting. The highest level briefly displays “Consumes quota faster,” fading back to the model name after about 1.6s. When the system has prefers-reduced-motion: reduce enabled, motion effects stop, and the scan freezes in the final state. Material colors are sampled from original Codex screenshots, representing a third-party aesthetic replication.
Lightning Switch and Subagent Preferences¶
After the lightning switch in the top left of the card is toggled on, it remains lit, and the state is persisted. It appends a “Prioritize using subagents” preference note to the end of the next message sent, without repeating for every message; when turned off, it appends a “Restore defaults” note instead. This is preference text; it does not guarantee the host will invoke subagents, nor can it bypass host routing restrictions.
Installation and Activation¶
In an environment with a configured DSH desktop profile, use the installation command from the profile:
dsh plugin --profile desktop add dsh-codex-power-slider
For local development, the package name can be replaced with link:C:\path\to\dsh-codex-power-slider (modify the path according to your local machine). The plugin loads client scripts with the permissions of the current dsh process; it is recommended to review the source code and MIT license on the GitHub repository before installing. package.json requires dsh.client.platform === "web", and the package name must match the insert.name in cordis.patch.yml, otherwise, it will be silently skipped.
Typical Usage¶
| Operation | Result |
|---|---|
| Click the model entry next to the input box | Opens the intensity card; the entry becomes a “Select Reasoning Intensity” pill |
| Click the model name in the card | Opens the grouped model list; hover to see Chinese intro |
| Drag slider / Click track | Switches reasoning intensity and writes to session (commit on release) |
| Click reset icon in top right | Returns to the model’s default level |
| Click lightning switch in top left | Appends a one-time subagent preference to the end of the next message |
Esc |
Model list → Card → Close |
The collapsed state has the same line height as the expanded state (approx. 28–30px), preventing composer reflow during switching. If the adapter provides only the effort id without a localized name, the plugin falls back to displaying names based on the mapping table in the README (e.g., max → “Deep”, ultra → “Ultra”); if effort.name exists, the adapter text is always prioritized.
Scenarios and Notes¶
Who it is for: Users who frequently adjust reasoning intensity within the DSH Web GUI and want an interface experience close to the Codex slider. The “Fun Customization” category also corresponds to its hand-written animations and zero-dependency single-file client implementation (modify lib/client.js and refresh; the host pushes rebuilt via SSE).
Limitations (Subject to README):
- Supports only the web platform; not for other DSH client platforms.
- In subagent routing sessions, the host forbids changing models and reasoning intensity: the plugin disables the slider and provides a Chinese prompt; this cannot be bypassed.
- If the catalog has no models, or the model has no reasoning metadata, the slider is not displayed.
- If the session is occupied by other DSH writing handles, the commit fails, and
session/writer-heldrelated error text is displayed in the card. - Mobile untested; Tested only on Chromium kernels, other engines untested.
Plugin read/write path overview: Read session.projections.faceOf('modelSelection'), list ctx.remote.session.modelCatalog(), write ctx.remote.session.selectModel({ sessionId, provider, model, reasoningEffort }). When writing DSH plugins, refer to PITFALLS.md in the repository.
Conclusion¶
dsh-codex-power-slider spreads the reasoning levels from the menu into a draggable Codex-style slider, combining it with a grouped model list, Ultra level feedback, and one-time subagent preference notes. It is suitable for scenarios where you want to adjust intensity long-term in DSH Web without occupying extra fixed UI space. See the SkillHub Plugin Page for catalog and details, and GitHub: NNNOOOMII/dsh-codex-power-slider for source code and full parameters.