Introduction¶
DeepSeek Harness (dsh) plugin: dsh-wrapped. Enter /wrapped to generate a shareable SVG summary card.
DSH sessions include real statistics, such as the time spent on tool calls versus thinking, streaming generation speed, and more. These details are usually shown only as small statistics bars and can be easily overlooked. dsh-wrapped turns these numbers into dark, screenshot-friendly cards for direct sharing.
Features¶
- Generate a shareable SVG summary card: Outputs a visual summary for display.
- Track rounds and steps: Records the basic scale of the session.
- Compare tool time and thinking time: Distinguishes model inference time from external tool-call time.
- Calculate average first-token generation time: Evaluates first-token response speed.
- Calculate decoding speed: Tracks generation rate in tokens/sec.
- Determine session type: Based on the ratio of tool time to thinking time, classifies the session as Explorer, Deep Thinker, or Balanced Operator.
Installation and Configuration¶
Install the plugin:
dsh plugin --profile <name> add dsh-wrapped
By default, the plugin does not take effect immediately. It must be added to the bundles list in the configuration file:
{
"dsh": {
"profile": {
"bundles": [
"@deepseek-ai/dsh-base",
"@deepseek-ai/dsh-web-app",
"dsh-wrapped"
]
}
}
}
To verify whether the configuration is active, use the dsh --profile <name> --dump-config command to check for the wrapped configuration item.
The optional configuration file is located at profiles/<name>/cordis.patch.yml:
- insert:
- id: wrapped
name: 'dsh-wrapped'
config:
minDecodeSampleMs: 1000
verdictRatioThreshold: 1.5
minDecodeSampleMs: Below this millisecond threshold (default 1000 ms), the decoding speed is displayed as—, avoiding data noise from too-small samples.verdictRatioThreshold: The threshold for the ratio of tool time to thinking time (default 1.5). If the ratio exceeds this value, the session is judged as Explorer or Deep Thinker; otherwise, it is Balanced Operator.
Usage¶
Enter /wrapped in any session.
After executing the command, it will:
1. Generate an SVG file at the path .dsh-wrapped/wrapped-<sessionId>-<timestamp>.svg.
2. Output a plain-text summary and the file path in the terminal.
3. Open the SVG file in a browser; then take a screenshot or convert it to PNG format for sharing.
Notes¶
- Dependencies: Requires the
@deepseek-ai/dsh-session-statsdependency (included by default in thedsh-web_appbundle). If this dependency is not installed, the command returns a clear error message. - Session requirement: The session must contain at least one completed step (
step/end); otherwise, a summary cannot be generated. - File path: The SVG file path is relative to the current working directory of the Harness process, not the project directory.
- Resource consumption: The command does not consume tokens, does not become part of the model history, and does not involve additional persistent data.
Conclusion¶
dsh-wrapped calculates statistics and renders SVGs through pure-function logic, without relying on the runtime environment, making its logic units testable. It leverages DSH’s own session projection data and requires no additional tracking mechanisms.
- Directory page: https://www.skillhub.cn/plugins/rand0wn/dsh-wrapped
- GitHub repository: https://github.com/rand0wn/dsh-wrapped