Introduction

The design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” In model inference and agent development scenarios, AI can quickly generate feature lists or PRDs, but it often records unverified assumptions as facts. This leaves product thinking at the documentation level, without convergence on risks and boundaries.

dsh-super-pm is a standard plugin for DSH used to package and distribute the Super PM product-thinking Skill. It does not provide a standalone GUI management system; instead, it integrates Super PM’s analysis methods, perspectives, and workflows into the DSH plugin ecosystem, helping developers progressively converge vague product ideas into verifiable and deliverable directions.

Plugin Positioning

This is a standard DSH plugin designed to address the problem of “treating AI assumptions as facts.” It is not simply a collection of prompts, but a plugin encapsulating the core capabilities of Super PM.

  • Author: Lohaslee
  • License: MIT License
  • Core features: supporting analysis for product decisions, decision accumulation, project state management, and delivery traceability.

Core Capabilities

The plugin provides a set of tools and methodologies focused on the quality of product thinking rather than pure text generation:

  1. Negative boundaries first: Before discussing features and solutions, first clarify what will not be done and what costs are unacceptable.
  2. Distinguish facts from assumptions: Record users’ original words, facts, inferences, suggestions, and assumptions pending validation separately, avoiding assertive wording that masks unknowns.
  3. Dynamically organize product perspectives: Select product dimensions dynamically based on the current issue, rather than mechanically rotating through all roles.
  4. Support for a Lead PM: One perspective can be designated as responsible for the overall direction, while other perspectives handle challenges and risk identification.
  5. Adapt to different product types: Support analysis of software, AI, hardware, services, content, and other product forms.
  6. Prioritize validation of high-risk assumptions: First propose the lowest-cost validation plan, then decide whether to perform competitive checks or prototype validation.
  7. Support decision accumulation: After explicit user authorization, save important decisions to the project’s .super-pm/decisions.md.
  8. On-demand artifacts: Can generate a decision brief, product report, or a PRD for an AI coding agent.
  9. Know when to step back: For pure engineering implementation, translation, or legal/finance conclusions, do not force a product perspective.
  10. Decision memory: Maintain consistency and historical traceability of product decisions within DSH sessions.
  11. Project state and validation: Track product stage, goals, owners, and validation records.
  12. Delivery and traceability: Support task board delivery drafts and traceability reports.
  13. PRD and sharing: Support PRD generation, validation, saving, and sanitized sharing of decisions.

Installation and Activation

The plugin can be installed with standard DSH plugin commands and supports installation from GitHub or a local directory.

Install from GitHub

Use the official command to install the latest version:

dsh plugin --profile web add github:Lohaslee/dsh-super-pm

If you need to pin a specific version for review:

dsh plugin --profile web add github:Lohaslee/dsh-super-pm#v0.8.3

Install from Local

If checking out the code locally:

dsh plugin --profile web add /absolute/path/to/dsh-super-pm

After installation, restart the dsh web service to load the new profile combination.

Verify Mounting

After installation, you can verify whether the plugin is mounted correctly using the configuration dump command. The output must include the following two lines; if any are missing, it is not acceptable:

dsh --profile web --dump-config | grep -A7 '^# == dsh-super-pm'

The output should include:

# == dsh-super-pm
- id: super-pm-tools
  name: dsh-super-pm
- id: super-pm-skill-filesystem
  name: '@deepseek-ai/dsh-skill-filesystem'

Usage

The plugin does not register a custom GUI itself; instead, it interacts through DSH tool interfaces and the filesystem. All tool and Skill calls require an explicit absolute project root directory.

1. Start a Product Request Directly

Enter a product-related request directly in DSH, for example:

“Use super-pm to help me diagnose the retention issue for this product”

2. Describe Decisions and Ideas

Directly describe product decisions, 0-to-1 ideas, feature definitions, or issues to diagnose; the plugin’s internal Skill will handle them.

3. Project File Operations

The plugin only reads and writes files under the project’s .super-pm/ directory. Save actions (such as saving decisions or saving a PRD) must be explicitly authorized by the user. When a direction changes, a change reason must be provided.

Notes

When using this plugin, keep the following facts in mind:

  1. Not a standalone system: It is not a standalone product management system, but a plugin that runs within DSH.
  2. No GUI registration: The current version does not register a custom GUI and relies on DSH’s native interfaces.
  3. No automatic research: The plugin does not automatically perform external research (such as scraping competitor materials); research must be initiated by the Skill or the user.
  4. File read/write permissions: The plugin only reads and writes the .super-pm/ folder inside the project.
  5. Save mechanism: Any save operation (PRD, Task Board, or decisions) requires presenting a draft first and obtaining user confirmation.
  6. Memory ownership: Product reasoning is handled by the Skill; the plugin is responsible only for project-level memory storage and tool distribution.
  7. Version compatibility: It depends on specific versions of @deepseek-ai/cordis, @deepseek-ai/dsh-skill, and @deepseek-ai/dsh-tools (>= 0.1.1-rc.2).

References

  • GitHub repository: https://github.com/Lohaslee/dsh-super-pm
  • Plugin directory: https://www.skillhub.cn/plugins/Lohaslee/dsh-super-pm
  • Upstream Skill: https://github.com/Lohaslee/super-pm