承上次定位:脚本生成空 skill 也能出 JSON,是因其 system 带写死的 _OUTPUT_PROTOCOL(进镜像、 永远在);提取没有这层兜底,只靠 skill 文件 → skill 一空就吐散文解析失败。本次对齐脚本做法: 给提取也加 _EXTRACT_OUTPUT_CONTRACT(写死的 entities/segments JSON 契约 + 不提商品等铁律), 拼在 skill 正文之后。即便将来 skill 再丢,提取也仍收到「只输出 JSON」硬指令,不会再退化。 真验证:用 dev 库里那条真实失败脚本,system 仅留写死契约(模拟 skill 全空)真调豆包 → 仍稳定出 JSON(角色/场景齐、商品不提)。 测试:152/152 全绿(新增契约兜底单测)。 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
AirShelf Backend
All backend code lives under AirShelf/core/backend by project decision.
Local bootstrap
cd /Users/maidong/Desktop/zyc/qiyuan_gitea/AirShelf/core/backend
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
python manage.py migrate
python manage.py runserver 0.0.0.0:8000
Start workers in separate terminals:
cd /Users/maidong/Desktop/zyc/qiyuan_gitea/AirShelf/core/backend
source .venv/bin/activate
celery -A airshelf worker -l info -P threads -c 4 # 线程池 4 并发(出图/视频可并行;勿用 -P solo 串行)
ffmpeg must be available on PATH for Stage5 export jobs.
Runtime layout
- Django project:
airshelf - Domain apps:
apps/* - Settings module:
airshelf.settings.development - Celery app:
airshelf.celery
Secrets must be supplied by environment variables or .env; never commit values from account.md.
Useful commands
python manage.py check
python manage.py makemigrations --check --dry-run
python manage.py migrate
python manage.py bootstrap_volcano_models
python manage.py test apps.accounts apps.projects apps.billing