跪拜 Guibai
← All articles
Agent · Artificial Intelligence

An Open-Source Agent Skill That Generates China Software Copyright Filings From Your Codebase

By IvanCodes ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Chinese software copyright registration is a bureaucratic bottleneck that forces developers to manually assemble application forms, operation manuals, and source-code extracts. This skill turns a multi-hour paperwork grind into a single agent prompt, with output that matches the format requirements of China's registration system.

Summary

The `software-certificate-skill` automates the tedious paperwork required for Chinese software copyright registration. Point it at a project directory, and it scans the codebase—configuration, routes, pages, components, services, APIs, models, and tests—to generate an application form, a formatted operation manual, and source-code extracts. It does not rely on README summaries; every generated function must trace back to actual code. The skill installs across multiple agent platforms including Codex, Claude Code, Cursor, and Qoder, and produces both editable DOCX and submission-ready PDF files. Operation manuals run 40–60 pages for typical projects, while code materials are capped at 60 pages split into front and back halves. Screenshot capture uses Playwright, and the pipeline includes post-conversion checks for pagination, images, and table integrity. The repository ships with unit tests, a validation script, and a self-test that runs against a sanitized project in a temp directory.

Takeaways
Scans project configuration, routes, pages, components, services, APIs, models, tests, and deployment files instead of relying on README descriptions.
Generates four deliverables: an application form info text file, an operation manual (DOCX/PDF), and source-code extracts split into front and back 30-page volumes.
Operation manuals use A4 black-and-white layout, typically 40–60 pages for substantial projects, and organize chapters around real project functions.
Code materials exclude dependencies, build artifacts, and lock files; source files retain original line numbers and a hash for traceability.
Supports automatic screenshot capture via Playwright, Computer Use, or manual upload; can generate a draft without screenshots.
Installs on Codex, Claude Code, Cursor, OpenCode, WorkBuddy, QoderWork, and TraeWork via a single Python script.
DOCX-to-PDF conversion includes post-checks for pagination, images, and tables; user must confirm ownership and dates.
Repository includes unit tests, a skill validation script, and a self-test that creates and cleans up a sanitized project in a temp directory.
Conclusions

Automating government paperwork is a practical stress-test for agent tooling: the output must conform to rigid bureaucratic formats, not just compile or pass tests.

By grounding every generated claim in actual code paths rather than documentation summaries, the skill avoids the hallucination risk that makes pure-LLM document generation unreliable for official filings.

The multi-platform install script signals that agent skills are becoming portable across IDEs, reducing lock-in to any single coding-agent ecosystem.

Concepts & terms
Software Copyright Registration (软著)
A Chinese government registration that records ownership of software. It requires submitting an application form, an operation manual, and source-code extracts in a specific format, and is often used to establish IP rights or meet policy requirements.
Agent Skill
A packaged set of instructions and tool integrations that an AI coding agent can invoke to perform a multi-step task—here, scanning a codebase and generating formatted documents—without the user needing to specify each step.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗