SIGN IN SIGN UP
langflow-ai / langflow UNCLAIMED

Langflow is a powerful tool for building and deploying AI-powered agents and workflows.

0 0 1 Python

feat(a2a): Phase 5b — address INPUT_REQUIRED wiring gaps

Fix the four gaps identified in Phase 5 audit:

Gap 1 (context plumbing): Pass A2A context (task_id, context_id,
task_manager, stream_bridge) through _execute_flow → simple_run_flow
→ graph.context. Agent components can now read self.ctx.a2a to detect
A2A execution. Actual Agent component hook is a separate lfx PR.

Gap 2+3 (streaming wiring): message:stream now passes A2A context
including stream_bridge reference. Checks for INPUT_REQUIRED state
after execution and emits the event to the SSE stream.

Gap 4 (end-to-end tests): Add 2 e2e tests where mock simple_run_flow
simulates the agent calling request_input mid-execution via
task_manager.request_input(). Tests verify the full round-trip:
send → INPUT_REQUIRED → poll shows question → follow-up resolves →
COMPLETED with client's answer in artifacts.

Also: document all gaps in a2a-implementation-plan.md audit section,
add _pending_inputs cleanup to test conftest.

124 total tests, all passing.
J
Jordan Frazier committed
fe044e3515feff5f4de917f4e7104fce1706d13a
Parent: 87d3697