L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

workflow-engine

workflow-engine

Operate on BiSheng's LangGraph workflow DAG engine, workflow nodes, graph state, callbacks, interruptions, and Celery workflow execution.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: workflow-engine description: "Operate on BiSheng's LangGraph workflow DAG engine, workflow nodes, graph state, callbacks, interruptions, and Celery workflow execution." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

workflow-engine

Use this sub-skill when a task touches BiSheng visual workflow execution: LangGraph DAG compilation, node registration, GraphState variable passing, conditional edges, callbacks, streaming events, input/output interruptions, workflow Celery tasks, or adding/debugging workflow nodes.

Start here

Run bundled helper commands from this sub-skill directory, or adjust the script path to this directory after import.

  1. Inspect node registration without importing the backend:
    python scripts/inspect_workflow_nodes.py --repo-root <bisheng-checkout>
    
  2. Read references/workflows.md for GraphEngine, node, callback, and worker workflows.
  3. Read references/troubleshooting.md when execution, resume, fan-in, callback, or Celery behavior is unclear.

Owned responsibilities

  • src/backend/bisheng/workflow/common/node.py node type definitions and node data schemas.
  • workflow/nodes/node_manage.py node factory and NODE_CLASS_MAP registration.
  • workflow/graph/graph_engine.py, graph_state.py, and workflow.py graph compilation and execution.
  • workflow/edges/edges.py conditional routing, fan-in, and branch handling.
  • workflow/callback/ callback events and LLM streaming bridges.
  • src/backend/bisheng/worker/workflow/tasks.py and Redis callback behavior for async execution, continuation, and stop signals.
  • Tests under src/backend/test/workflow/ and workflow-related API/service tests.

Route sibling areas instead of duplicating them

  • Use frontend-apps for React workflow canvas UI, node panels, route guards, or request wrappers.
  • Use knowledge-rag for KnowledgeRetriever, QA retriever, RAG pipeline internals, and retrieval quality.
  • Use linsight-mcp for Linsight autonomous task-mode workflows, SOP/Skill runtime, and MCP tools.
  • Use identity-permissions-tenancy for workflow app visibility, app permissions, tenant filters, and cursor list permission scans.
  • Use backend-core for route/envelope/service-layer changes not specific to workflow execution.

Non-negotiables

  • Add a workflow node by updating the enum, node class, and factory map together.
  • Preserve GraphState variable reference semantics ({node_id}.{variable_key} and indexed suffixes) when changing node outputs.
  • Respect INPUT and interactive OUTPUT interruption/resume behavior; do not bypass LangGraph checkpoints casually.
  • Keep callback event payloads compatible with frontend streaming and persisted message expectations.
  • Do not run long workflow executions, external tools, or LLM calls as verification unless the environment and credentials are explicitly approved.
Skills enregistrées