Introduction¶
The philosophy of DeepSeek Harness (DSH) is “everything is a plugin”. For DSH or agent developers, daily development often encounters several types of problems: difficulty in exporting and archiving conversation records, inconvenience in passing long task context across conversations, lack of unified management for API call costs, insufficient directness in retrieving historical conversations, and the need for scattered handling of Prompt tuning, multi-model comparison, task orchestration, and security auditing.
dsh-companion-dev is the official companion plugin for DeepSeek Harness, maintained by beijingwahw, built on the Cordis framework and Harness Plugin SDK. It provides nine major capabilities: intelligent conversation export, context handoff summaries, API cost optimization, global conversation search, execution trace analysis, Prompt engineering workbench, multi-model arena, task orchestration, and security auditing. The following introduces its positioning, installation methods, typical usage, and notes.
What is it¶
- Repository:
beijingwahw/dsh-companion-dev - License:
MIT - Positioning: Official companion plugin for DeepSeek Harness; Developer Edition includes the nine modules listed in the documentation.
- Lite Version: If you only need the four basic modules A–D, refer to
beijingwahw/dsh-companion. - Runtime Framework: DeepSeek Harness, Cordis ≥ 4.0
- Development Language: TypeScript,
strict: true, ESM - Prerequisites: DeepSeek Harness
>= 0.1.0; Node.js^22.19 || >=24 - Data Location: User data is stored only locally within the Harness plugin sandbox.
- Key Storage: API Keys are encrypted to disk using
AES-256-GCM. - Package Name:
dsh-companion; manifest id isdeepseek-companion. - Module Relationship: The nine modules are all independent Cordis sub-plugins; they can be started and stopped arbitrarily in any combination without affecting each other.
The function numbers listed in the documentation are A, B, C, D, E, F, G, H, J.
Core Features¶
A · Intelligent Conversation Export¶
Supports exporting conversations to Markdown, PDF, JSON, or PNG long images; provides timestamp toggles and privacy masking; also supports batch ZIP packaging of multiple conversations.
B · Context Handoff Summary¶
Supports generating four-paragraph summaries of no more than 500 characters; allows editing, copying to clipboard, and saving as templates. Summaries can also be imported into new conversations as the first message of the system role for context handoff.
C · API Cost Optimization¶
Provides encrypted hosting of API Keys, official dynamic pricing, peak-valley time-of-use pricing, peak-valley automatic scheduling, intelligent model routing, daily/monthly tiered budgets, and token consumption and cost reports.
D · Global Conversation Search + In-Chat Search¶
Supports keyword fuzzy search, time range filtering, and tag filtering; search results can jump directly to the conversation. A floating find bar is also provided within the conversation.
E · Execution Trace Analyzer¶
Provides a trace timeline, allowing you to identify the 3 slowest and most expensive steps; supports anomaly annotation, diffing two traces, daily aggregated trends, and baseline alerts.
F · Prompt Engineering Workbench¶
Provides Prompt version management, A/B testing, a template library, and batch validation of structured output JSON Schemas.
G · Multi-Model Arena¶
Supports parallel comparison of multiple models using the same Prompt, batch evaluation leaderboards, a model recommendation engine, and encrypted hosting of external vendor keys.
H · Breakpoint Resume and Task Orchestration¶
Provides a visual pipeline, breakpoint resume, batch task queues, and Cron/Chinese natural language scheduling.
J · Security and Auditing¶
Provides multi-API Key management, operation audit logs, DLP, and compliance reports.
Installation and Enablement¶
Installation¶
It is recommended to install using the dsh CLI. --profile is used to specify the target profile, for example web:
dsh plugin add beijingwahw/dsh-companion-dev --profile web
dsh web
If you wish to install from source, you can clone the repository and build first:
git clone https://github.com/beijingwahw/dsh-companion-dev.git
cd dsh-companion-dev
pnpm install
pnpm run build
Verify Installation¶
Execute usage in the Harness command palette. If it returns the monthly usage text report, the plugin has been mounted correctly.
Configure Module C¶
If you want to use API cost optimization capabilities, first perform these two steps:
- Turn on the “Developer Mode” master switch in the plugin settings page.
- Enter and save the DeepSeek API Key in “API Key Management”; testing the connection is optional.
Typical Usage¶
Export Conversation¶
Click “Export” in the header operation area of the conversation interface, or execute export in the command palette. Then select Markdown, PDF, JSON, or PNG long image. Multiple conversations can also be exported in batches as ZIP.
Generate Handoff Summary¶
Click “Handoff Summary” at the top of the conversation, or execute the command handoff. After generation, you can copy it to the clipboard or save it as a template. When creating a new conversation, you can use handoff-import to import historical summaries.
Retrieve Historical Conversations¶
Use the global search box at the top of the historical conversation list, or execute the command search. After entering a specific conversation, you can press Ctrl/Cmd+F to open the floating find bar within the conversation.
View Usage¶
Execute usage in the command palette to view the monthly usage text report.
Module Toggles¶
The plugin root configuration contains the following toggles:
enableExport
enableHandoff
enableCost
enableSearch
enableTrace
enablePrompt
enableArena
enableOrchestrator
enableSecurity
apiBaseUrl
apiTimeoutMs
The first nine toggles default to true. The default value for apiBaseUrl is:
https://api.deepseek.com
The default value for apiTimeoutMs is:
60000
Applicable Scenarios and Notes¶
This plugin is suitable for developers using DeepSeek Harness, especially those who need to handle the following:
- Exporting conversation records or archiving conversation content
- Passing long task context to new conversations for continued processing
- Managing API Keys, budgets, and call costs
- Retrieving historical conversations and conversation content
- Analyzing execution traces
- Managing Prompt versions and performing A/B testing
- Comparing multiple models
- Orchestration tasks, breakpoint resume, and scheduled scheduling
- Viewing security audit and compliance information
Pre-use notes:
- The plugin runs with the permissions of the current
dshprocess; check the source code and license before installing. - This repository is the Developer Edition; if you only need basic capabilities A–D, refer to
beijingwahw/dsh-companion. - Hot Module Replacement (HMR) during development requires Node
>= 24.11; internal interfaces in early 24.x Node versions are incompatible withcordis-plugin-loader 1.0.2. - The nine modules can be started and stopped independently; disabling one module will not affect the others.
- API Keys are encrypted to disk using
AES-256-GCM, and user data is stored only locally within the Harness plugin sandbox.
Conclusion¶
dsh-companion-dev integrates conversation export, context handoff, cost optimization, search, trace analysis, Prompt engineering, multi-model comparison, task orchestration, and security auditing into the DeepSeek Harness plugin ecosystem. If you are using DSH for agent development and want to handle these engineering requirements within a single plugin suite, you can install and verify it using the commands above.