Preface¶

When taking over a project with DeepSeek Harness (DSH) or other AI programming assistants, a common issue is: the model is unfamiliar with the project’s tech stack, directory conventions, and testing strategies, leading to inconsistent code styles and frequent CI failures. Manually writing AGENTS.md, documentation skeletons, and pipelines can take half a day to a full day for each new project.

Project Blueprint is a workflow-type DSH plugin maintained by community contributor shuguang1994 (current version v1.7.1, MIT license, GitHub 13 stars). Instead of filling static templates, it scans project files, identifies dependencies and tech stacks, and dynamically assembles specification documents and engineering infrastructure. Below, we introduce its positioning, capabilities, and usage.

What Is This¶

Project Blueprint is a reusable AI agent skill packaged as a DSH plugin. With a single command, it can equip a new project (or incrementally enhance an existing one) with the specification system needed for AI development.

Unlike manually writing AGENTS.md or applying fixed scaffolds, it uses an autonomous discovery engine: scanning 30+ file patterns, classifying dependencies through three layers (precise knowledge base matching → 29 heuristic rules → web search fallback), and dynamically assembling outputs from a knowledge base of 70+ components. This ensures the generated content reflects the project’s actual tech stack.

Maintainer: shuguang1994
Category: Workflow
Directory Page: https://www.skillhub.cn/plugins/shuguang1994/project-blueprint
Source Code: https://github.com/shuguang1994/project-blueprint

Core Features¶

Autonomous Discovery and Classification¶

The plugin does not rely on predefined file lists; instead, it automatically analyzes after scanning the project:

• Project structure: monorepo, front-end/back-end separation, monolithic project
• Business type: 13 heuristic inferences
• Dependency classification: knowledge base hits, name pattern matching, real-time web search for unknown dependencies

Tech stack coverage (from README):

Generated Outputs¶

One initialization can produce the following files (existing files are not overwritten in incremental mode):

Other Capabilities¶

• Incremental Mode: For existing projects, only gaps are filled without overwriting current files.
• Multi-IDE Support: Beyond DSH, it is compatible with 28+ agent environments like Claude Code, Cursor, GitHub Copilot, Codex, Windsurf, etc.
• Self-Evolving Rules: The generated AGENTS.md includes auto-maintenance conventions, updating module tables and tech stacks as the project grows.
• MCP Tool Recommendations: Generates tool lists and combination suggestions based on the detected tech stack.

Installation and Activation¶

DSH plugins are distributed via the SkillHub community directory. SkillHub is an independent community site with no official affiliation to DeepSeek / High-Flyer.

Before installation, it is recommended to review the source code and MIT license to confirm acceptance of the plugin reading and writing project files with the current DSH process permissions.

Install in DSH (official README command):

dsh plugin --profile web add 'github:shuguang1994/project-blueprint'

If installing via the skills.sh ecosystem (supports DSH and other agents):

# Global (GitHub)
npx skills add shuguang1994/project-blueprint

# Domestic (Gitee mirror, no proxy needed)
npx skills add https://gitee.com/shuguang1994/project-blueprint.git

# Subsequent updates
npx skills update project-blueprint

Typical Usage¶

After installing and enabling the plugin, simply issue an initialization command to the target project in a DSH session. The README provides an example:

Initialize this project's development standards

In Chinese, you can express the same intent, such as “初始化这个项目的开发规范”. The agent will sequentially execute file scanning, dependency classification, and documentation and pipeline generation. For projects with existing partial specifications, the incremental mode only fills in missing items.

If the project contains third-party libraries not listed in the README, the plugin will fetch conventions via web search; the generated AGENTS.md also requires verifying APIs and versions against official documentation before coding.

Applicable Scenarios and Notes¶

Suitable For:

• New projects that need to quickly establish AGENTS.md, documentation skeletons, CI, and testing strategies.
• Existing projects with incomplete specifications, aiming to fill gaps incrementally.
• Projects with tech stacks within the plugin’s knowledge base coverage (7 languages, 15 frameworks, and 70+ components), or those fallback-capable via web search.

Notes:

• The plugin runs with DSH process permissions and can read/write workspace files; review the source code before installation.
• Generated specifications are a starting point; complex business logic and architectural decisions still require manual review.
• .husky/pre-commit is only for JS/TS projects; other languages should be adjusted as needed.
• Industry data referenced in the README (such as the adoption scale of AGENTS.md in open-source repositories and the reduction in error rewrites in Anthropic benchmarks) comes from the plugin documentation and is not independently verified in this article.

Summary¶

Project Blueprint consolidates “scan project → identify tech stack → generate specifications and engineering infrastructure” into a single command, suitable for equipping new or existing projects with AI development infrastructure in DSH workflows. For details and updates, refer to the directory page and GitHub repository:

• Directory Page: https://www.skillhub.cn/plugins/shuguang1994/project-blueprint
• GitHub: https://github.com/shuguang1994/project-blueprint