AIChain Skills

Agent skill for AIChain developer integration. This skill helps coding assistants gather the minimum AIChain-specific requirements, generate an integration profile, implement WebAPI/SDK code, and run real-service smoke tests with bundled text, audio, and image fixtures.

Installation

Install this repository with any Agent Skills compatible installer:

npx skills add <github-owner>/<github-repo>

For this local checkout, the skill entrypoint is:

SKILL.md

This repository is a standard skill package. It intentionally does not include a custom CLI or npm package wrapper.

Available Skills

Skill	Description
aichain-integration	Integrate AIChain through WebAPI, JavaScript SDK, Python SDK, or Android SDK with guided profile collection and smoke testing.

Configuration

Real smoke tests require AIChain credentials:

export AICHAIN_ENDPOINT=""
export AICHAIN_APP_ID=""
export AICHAIN_API_KEY=""

AICHAIN_ENDPOINT must match AICHAIN_REGION. For example, use the cn endpoint with AICHAIN_REGION="cn" and the us endpoint with AICHAIN_REGION="us".

Optional configuration:

export AICHAIN_REGION="cn"
export AICHAIN_MODEL_ID="b16924f42ffe4fd895d4ba4778278bc3"
export AICHAIN_CHUNK_TTS="true"
export AICHAIN_PROFILE='{"region":"cn","modelId":"b16924f42ffe4fd895d4ba4778278bc3","capabilities":"nlu"}'
export AICHAIN_JS_SDK_PACKAGE=""
export AICHAIN_PYTHON_SDK_PACKAGE=""

Defaults:

• region: cn; use us or another region code when the client is deployed outside China, with the matching endpoint
• modelId: b16924f42ffe4fd895d4ba4778278bc3
• Pure acoustic VAD: minSilenceDuration=600ms
• Acoustic + semantic VAD: minSilenceDuration=300ms, minEndpointingDelay=250ms
• chunk_tts: enabled when the developer is unsure and TTS is enabled

Integration Support

The skill supports:

• WebAPI / WebSocket
• JavaScript / TypeScript SDK
• Python SDK
• Android SDK

Before coding, the skill asks focused multiple-choice questions for client region, capability mix, ASR language, duplex/VAD mode, interruption strategy, image understanding, audio codec, optional chunk_tts, and integration path. Every option includes a short explanation.

References

Reference files are loaded only when needed:

• references/webapi.md
• references/js-sdk.md
• references/python-sdk.md
• references/android-sdk.md
• references/testing.md

The references keep links to the official Yuque docs. Exact endpoint, auth, event, SDK method, and parameter names should be checked against the official docs before production code generation.

Smoke Tests

Run the path that matches the selected integration:

node scripts/smoke-webapi.mjs
node scripts/smoke-js.mjs
python3 scripts/smoke-python.py

Bundled fixtures:

• Chinese STT audio: assets/audio/zh_sample_pcm.raw
• Chinese expected text: assets/audio/zh_expected.txt
• English STT audio: assets/audio/en_sample_wav.wav
• English expected text: assets/audio/en_expected.txt
• Image fixture: assets/image/sample.png
• Image question: assets/image/expected_question.txt
• Image expected answer substring: assets/image/expected_answer_contains.txt
• Text fixture: assets/text/sample.txt

The scripts fail explicitly when required credentials are missing. STT checks validate Chinese and English expected terms; image understanding asks 手里是什么？ and expects an answer containing 苹果.

License

MIT