cangtianhuang/humanizer-academic-zh

1. Repository type

- This is a prompt artifact repository, not an application codebase.

- Its “architecture” is mostly about packaging and distributing a single behavioral policy for LLMs: a Chinese academic text humanizer.

2. Core purpose

- The repository provides instructions that tell an LLM how to reduce AI-like writing signals without changing meaning.

- The design goal is explicitly short, precise, and token-efficient, so the payload is minimal and intended for direct insertion into model context.

3. Main functional subsystem

- The only meaningful subsystem is the prompt ruleset.

- Those rules are designed around surface-pattern rewriting: preserve content, adjust wording/style to look more human and academically natural.

4. Two packaging formats for the same behavior

- SKILL.md is the Claude Code Skills version, with YAML frontmatter so it can be installed as a skill.

- SYSTEM_PROMPT.md is the generic system-prompt version for any LLM.

- Both files are described as having the same core content, just different delivery mechanisms.

5. Documentation and entry point

- README.md is the human-facing control plane of the repo.

- It explains what the skill does, how to install it, and how to invoke it in either Claude Code Skills mode or raw system-prompt mode.

- LICENSE is operationally separate and only governs reuse.

6. Input → processing → output flow

- Input: a user pastes a Chinese academic paragraph or paper fragment.

- Processing: the LLM applies the prompt rules to remove AI-like traces while maintaining meaning.

- Output: a rewritten academic-style text segment.

- No intermediate services, databases, or pipelines exist in the repository itself; the transformation happens entirely inside the consuming LLM.

7. Distribution and activation boundary

- The repo supports two activation paths:

- Claude Code Skills installation via SKILL.md

- Direct system prompt insertion via SYSTEM_PROMPT.md

- This creates a clean boundary between authoring the prompt and executing it in a model client.

8. External tooling that matters

- The README references npx skills as an external distribution/installation tool.

- It also references downstream consumers such as Claude Code, Cursor, Codex, ChatGPT, DeepSeek, etc.

- These are not part of the repo’s runtime, but they are the real deployment targets for the prompt.

9. What is intentionally absent

- There is no backend service, UI, API server, build system, or test suite.

- There is also no dataset, model weights, or retrieval layer.

- Architecturally, this is a static prompt package with documentation.

10. How to diagram it

- Draw the repo as a source-of-truth prompt bundle with two output artifacts: SKILL.md and SYSTEM_PROMPT.md.

- Show README.md as the installer/instruction layer.

- Show the consumer side as an LLM client/runtime that receives the prompt and a user’s text, then emits polished academic Chinese.