backend-api
Operate TaskingAI backend REST/OpenAI-compatible APIs, object lifecycle, assistant generation, retrieval objects, auth, files, and backend service troubleshooting.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- backend-api
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate TaskingAI backend REST/OpenAI-compatible APIs, object lifecycle, assistant generation, retrieval objects, auth, files, and backend service troubleshooting.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/api-and-object-model.md17.0 KB
- 参考文档references/native-testing.md5.5 KB
- 参考文档references/troubleshooting.md10.4 KB
03 SKILL.md 正文
backend-api
Use this sub-skill when the task is to work with TaskingAI's backend API/web service semantics: REST objects, lifecycle order, assistant/chat/message generation, retrieval collections/records/chunks, models, actions, bundle instances as backend tool references, OpenAI-compatible endpoints, authentication, file/image uploads, backend dependencies, or backend-native test prerequisites.
Load this first
- For route prefixes, auth mode, object lifecycle, request shapes, generation order, retrieval order, and OpenAI-compatible mappings, read API and object model.
- Before proposing or interpreting backend tests, read native testing for Python-version constraints, DB/Redis/service prerequisites, and safe skip criteria.
- When a backend service, import, generation, retrieval, auth, file, or storage symptom appears, use troubleshooting.
Source-backed facts to preserve
- Backend package version evidence is
v0.3.0. - The backend builds a FastAPI app with an
APIRouter; API mode uses/v1, while web-console mode uses/api/v1. - Verified backend imports require Python 3.10. Python 3.11 backend import fails because
aioredis==2.0.1defines a duplicateTimeoutErrorbase class under Python 3.11. - Verified model type values are
chat_completion,text_embedding,rerank, andwildcard. - Verified retrieval splitting includes
TextSplitterwithtokenandseparatormodes.
Quick routing
- Starting containers, selecting Docker/Compose topology, setting environment variables, ports, storage mode, and service URL wiring belong to
../deployment-configuration/. - Provider-specific model schema catalogs, credentials, inference-provider behavior, and provider execution failures belong to
../inference-providers/. - Plugin bundle catalog internals, plugin schemas, and plugin-service execution internals belong to
../plugin-bundles/. - Backend-created action objects, bundle instances as backend tool references, assistant tool wiring, and tool execution from an assistant generation flow stay here; return to plugin/provider sub-skills only for the provider or bundle-specific payload details.
Operating workflow
- Identify service purpose and auth. API mode expects API-key bearer auth and
/v1routes. Web mode expects admin bearer auth and/api/v1routes. Do not mix route prefixes with the wrong auth mode. - Plan object order before calling generation. For assistant + retrieval + tools, create or select required models first, then retrieval collections/records/chunks, actions or bundle instances, assistant, chat, user message, and finally generation. The exact order and constraints are in API and object model.
- Validate IDs and model capabilities. Backend route validation uses path IDs, object parents, model types, and model properties. Chat models must be
chat_completion; retrieval collection models must betext_embedding; rerank is optional for query reranking; function-call tools require a chat model that supports function calling. - Treat generation as orchestration. Stateful generation loads chat memory, retrievals, tools, and model settings, then may run retrieval/tool function-call rounds before creating the assistant message.
- Check backend dependencies early. DB/Redis/object storage/inference/plugin service availability affects imports, app startup, object writes, retrieval embedding, file/image uploads, and generation. Use native testing and troubleshooting to decide whether a failure is an app bug, missing service, wrong Python, or unavailable external dependency.
Hard cases this sub-skill covers
- Mapping a request like "assistant with RAG over one document and a tool call" into the required model → collection → record/chunk → action/bundle-instance → assistant → chat → user message → generation sequence, with the validation points that can fail at each step.
- Explaining why backend import or inspection under Python 3.11 fails and choosing Python 3.10 for backend inspection/development instead of debugging unrelated FastAPI or Pydantic code.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.