bisheng
Use BiSheng's enterprise LLM application DevOps platform repository, including FastAPI backend, workflow/RAG/Linsight runtimes, permissions, dual React frontends, deployment, and maintainer workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- bisheng
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use BiSheng's enterprise LLM application DevOps platform repository, including FastAPI backend, workflow/RAG/Linsight runtimes, permissions, dual React frontends, deployment, and maintainer workflows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/architecture-map.md5.0 KB
- 参考文档references/repo-provenance.md3.0 KB
- 参考文档references/repo-routing-metadata.json487 B
- 参考文档references/troubleshooting.md5.0 KB
- 脚本scripts/check_repo_surface.py2.8 KB
03 SKILL.md 正文
BiSheng
Use this repo skill when a task names BiSheng, 毕昇, dataelement/bisheng, or asks about this enterprise LLM application DevOps platform's backend, workflow engine, knowledge/RAG pipeline, Linsight agent, permissions/tenancy, dual React frontends, Docker deployment, or repository maintenance.
This root skill is a router. Read the focused sub-skill first, then return here for cross-cutting checks.
First checks
- If you are checking whether this skill matches a checkout, read repo provenance.
- For the monorepo and runtime topology, read architecture map.
references/repo-routing-metadata.jsonis structured metadata for managed repo-skill routing and import checks.- For cross-cutting install/config/runtime failures, read troubleshooting.
- For a safe checkout inventory, run the bundled helper against a BiSheng checkout:
python scripts/check_repo_surface.py --repo-root <bisheng-checkout>
Install and verify a checkout
For backend work, install from the backend root with the repository-supported lockfile:
cd <bisheng-checkout>/src/backend
uv sync --frozen
uv run python -c "import bisheng, bisheng_langchain; print(getattr(bisheng, '__version__', 'unknown'))"
For frontend work, install and test from the owning app directory only:
cd <bisheng-checkout>/src/frontend/platform && npm install && npm test
cd <bisheng-checkout>/src/frontend/client && npm install && npm run test:ci
Use narrower sub-skill commands before running broad suites; many integration/e2e paths require MySQL, Redis, Milvus, Elasticsearch, MinIO, OpenFGA, or the commercial gateway.
Route by task
| User task | Read |
|---|---|
| Add or debug FastAPI routes, DDD modules, response envelopes, settings/context, SQLModel DAO/model behavior, backend tests, or error-code handling | backend-core |
| Work on workflow canvas execution, LangGraph graph compilation, node types, callbacks, interruption/resume, workflow Celery tasks, or new workflow nodes | workflow-engine |
| Work on knowledge libraries/spaces, file upload states, document parsing, Load → Transform → Ingest, Milvus + Elasticsearch recall, MinIO artifacts, QA libraries, or knowledge workers | knowledge-rag |
| Work on Linsight task mode, SOP/Skill migration, Redis worker queue, task/session state events, deepagents runtime, built-in tools, or MCP integrations | linsight-mcp |
| Work on JWT auth, RBAC/ReBAC/OpenFGA, PermissionService, tenant isolation, admin scope, approval center, org/SSO sync, commercial gateway, or cursor-pagination permission performance | identity-permissions-tenancy |
| Modify Platform or Client React apps, routes, stores, request wrappers, UI systems, i18n, brand theme, tests, or Vite config | frontend-apps |
| Install, run, deploy, configure, migrate, operate, or maintain the repo, Docker Compose stack, backend scripts, arch guard, uv/npm commands, or SDD workflow | deployment-maintenance |
Cross-route decisions
- Backend versus frontend: API contract or permission semantics belong to backend/identity sub-skills; rendering, request wrappers, route guards, and state belong to
frontend-apps. - Workflow versus Linsight: visual flow DAGs and workflow nodes belong to
workflow-engine; autonomous task-mode execution, SOP/Skill runtime, Redis Linsight worker, and MCP tool wrapping belong tolinsight-mcp. - Knowledge versus permissions: document ingestion/retrieval belongs to
knowledge-rag; resource visibility, ReBAC checks, approval modes, tenant filters, and cursor permission performance belong toidentity-permissions-tenancy. - Deployment versus code change: runtime service wiring, Docker, config layering, migrations, and operational scripts belong to
deployment-maintenance; feature implementation details route to the owning code sub-skill.
Non-negotiable repository laws
- Backend layering is Router → Endpoint → Service → Repository/DAO → DB. Do not bypass it for new work.
- Every resource authorization path goes through PermissionService/OpenFGA-aware services; do not query role-access tables as a new permission shortcut.
- Multi-tenancy is ContextVar-driven. ORM SELECTs are auto-filtered, but raw SQL and bulk update/delete need explicit tenant-safe design.
- MySQL and DM8 compatibility are both mandatory. Use dialect helpers for JSON, large text, timestamp defaults, and keyset predicates.
- The two frontends are separate SPAs. Do not mix Platform's Zustand/react-query v3/bs-ui stack with Client's Recoil/TanStack Query v4/shadcn stack.
- Never import
axiosdirectly in business frontend code; use the app's wrapped request layer. - Do not write plaintext secrets into config files. Password fields in BiSheng config are Fernet-encrypted.
Verification and helper scope
The bundled scripts are read-only inspectors. They accept --repo-root so they can be used with any BiSheng checkout; they do not depend on the checkout used to create this skill.
Use repository-native tests only after selecting the focused surface. Many end-to-end and integration tests require MySQL, Redis, Milvus, Elasticsearch, MinIO, OpenFGA, frontend browsers, or the commercial gateway; prefer the narrow safe test listed by the owning sub-skill before running broad suites.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).