Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.
Python agentagentic-aiagentic-frameworkagentic-workflowaiautomationclaudedeepseekgenaigptllmlow-codemcpnextjsno-codeopenaipythonskillsworkflow
"""Config pre-pulls run inside Shell and persist only completed initialization."""
import json
import httpx
import pytest
from dify_agent.adapters.shell.protocols import CompleteShellCommandResult
from dify_agent.layers.config import layer as config_module
from dify_agent.layers.shell.layer import ShellSession
from dify_agent.protocol.schemas import RunLayerSpec
from dify_agent.runtime.context import Services
from dify_agent.runtime_backend import RuntimeBackendProfile
from tests.local.dify_agent.runtime.test_runner import Backend, Commands, request, run_runner, runtime_specs
@pytest.mark.anyio
async def test_first_pull_failure_retries_then_current_config_refreshes_without_repull(monkeypatch):
commands = Commands()
backend = Backend(commands)
pulls = []
fail = True
async def pull(session, script, **kwargs):
assert backend.active
assert kwargs["inject_agent_stub_env"] is True
pulls.append(script)
output = (
""
if fail
else json.dumps(
{"items": [{"name": "skill", "directory_path": "/skills/skill", "skill_md": "Instructions from skill"}]}
)
)
return CompleteShellCommandResult("pull", "exited", True, 1 if fail else 0, output, True, None, len(output))
monkeypatch.setattr(ShellSession, "run_remote_script", pull)
seen_instructions = []
async def stream(messages, info):
assert "Instructions from skill" in info.instructions
seen_instructions.append(info.instructions)
yield "done"
def spec(note, *, writable=False):
return RunLayerSpec(
name="config",
config={
"mentioned_skill_names": ["skill"],
"note": note,
"config_version": {"kind": "build_draft" if writable else "snapshot", "writable": writable},
},
)
async with httpx.AsyncClient() as client:
services = Services(client, client, runtime_backend_profile=RuntimeBackendProfile(None, backend))
# Config precedes both resources in the submitted list; native ordering owns entry.
first, _ = run_runner(request(spec("original"), *runtime_specs()), services, monkeypatch, stream)
with pytest.raises(config_module.DifyConfigLayerError):
await first.run()
assert first.terminal_session_snapshot.layers["config"]["initialized"] is False
assert backend.released == 1
fail = False
second, _ = run_runner(
request(spec("original"), *runtime_specs(), snapshot=first.terminal_session_snapshot),
services,
monkeypatch,
stream,
run_id="second",
)
await second.run()
assert second.terminal_session_snapshot.layers["config"]["initialized"] is True
third, _ = run_runner(
request(spec("revised", writable=True), *runtime_specs(), snapshot=second.terminal_session_snapshot),
services,
monkeypatch,
stream,
run_id="third",
)
await third.run()
assert len(pulls) == 2
assert backend.acquired == backend.released == 3
state = third.terminal_session_snapshot.layers["config"]
assert "original" in seen_instructions[0] and "revised" in seen_instructions[1]
assert "dify-agent config env push --help" not in seen_instructions[0]
assert "dify-agent config env push --help" in seen_instructions[1]
assert "dify-agent config prompt pull --help" in seen_instructions[0]
assert "dify-agent config prompt pull --help" in seen_instructions[1]
assert "dify-agent config prompt push --help" not in seen_instructions[0]
assert "dify-agent config prompt push --help" in seen_instructions[1]
assert "already-built run requests keep their prompt" in seen_instructions[1]
assert set(state) == {"initialized", "pulled_skill_outputs", "pulled_file_outputs"}
assert "Instructions from skill" in state["pulled_skill_outputs"]["skill"]