Jiaoquaner AI Text Humanizer

Problem

When reviewing LLM-generated copy, the issue is often not whether text can be produced, but whether it sounds too machine-like: repetitive structures, overconfident phrasing, and a template-like tone. The Jiaoquaner reduce-AI-rate API turns this editing step into a synchronous request: send the original text and receive a full rewritten version, which is useful for adding humanization to an existing content pipeline.

How It Works

The skill wraps scripts/reduce_ai_rate.sh and reduce_ai_rate.ps1 with the same parameters, output conventions, and exit codes. It reads JIAOQUANER_API_KEY or jiaoquaner_api_key from the environment, then handles escaping for quotes, newlines, Chinese characters, emoji, and backslashes so the caller does not hand-build JSON strings. After the request is sent to the Jiaoquaner reduce-ai-rate endpoint, the script treats code == 0 as success, writes the rewritten text to stdout, and sends metadata such as content_id or error hints to stderr.

Key points:
- On success: read stdout directly and do not merge stderr into the final body.
- On failure: follow the exit code, covering cases such as insufficient balance, invalid API key, prohibited content, or internal service errors.
- For long text: prefer file-based input and output flags instead of placing the whole document on the command line.
- HTTP 200: is not enough; the API code field is the source of truth.

Boundaries

This is a single-call, synchronous rewriting tool. It does not perform local AI-rate detection, automatic retries, or batch and concurrent processing. Every successful call is billed, so avoid test-only runs and do not issue multiple calls for multiple segments without first telling the user that it will incur repeated charges. It fits manual or semi-automated review workflows with an existing API key, not offline local rewriting or fully automatic retry pipelines.