Merge pull request #1619 from CREDO23/feat-run-citations

[Feat] Cite scraper runs as verifiable sources in chat
This commit is contained in:
Rohan Verma 2026-07-22 14:53:45 -07:00 committed by GitHub
commit ca4f231577
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 339 additions and 30 deletions

View file

@ -37,6 +37,18 @@ def test_web_result_maps_to_url() -> None:
assert to_frontend_payload(entry) == "https://example.com/a"
def test_run_maps_to_run_handle() -> None:
entry = _entry(CitationSourceType.RUN, {"run_id": "run_abc-123"})
assert to_frontend_payload(entry) == "run_abc-123"
def test_run_without_handle_is_dropped() -> None:
entry = _entry(CitationSourceType.RUN, {})
assert to_frontend_payload(entry) is None
def test_not_yet_renderable_kind_is_dropped() -> None:
entry = _entry(CitationSourceType.CHAT_TURN, {"thread_id": 1, "turn": 2})

View file

@ -69,6 +69,12 @@ def _verb_tool(tools, name: str):
return next(t for t in tools if t.name == name)
def _invoke(tool, text: str, *, state=None):
"""Call the coroutine with a stand-in runtime (ToolNode injects it in prod)."""
runtime = SimpleNamespace(state=state or {}, tool_call_id="tc_1", context=None)
return tool.coroutine(runtime, text=text)
async def test_registry_becomes_one_tool_per_verb_plus_readers(isolate):
caps = [
_capability(name="web.scrape", output=_EchoOutput(echoed="a")),
@ -104,20 +110,51 @@ async def test_tool_runs_executor_and_returns_serialized_output(isolate):
tools = isolate.module.build_capability_tools(workspace_id=7, capabilities=[cap])
tool = _verb_tool(tools, "web_scrape")
result = await tool.ainvoke({"text": "ping"})
result = await _invoke(tool, "ping")
# Fake session makes record_run fail -> no run_id key, plain serialized output.
assert result == {"echoed": "hi there"}
assert cap.executor.seen.text == "ping"
async def test_tool_registers_run_citation_when_stored(isolate, monkeypatch):
from langgraph.types import Command
cap = _capability(name="web.scrape", output=_EchoOutput(echoed="hi"))
monkeypatch.setattr(isolate.module, "record_run", AsyncMock(return_value="abc-123"))
tools = isolate.module.build_capability_tools(workspace_id=7, capabilities=[cap])
tool = _verb_tool(tools, "web_scrape")
result = await _invoke(tool, "ping")
assert isinstance(result, Command)
registry = result.update["citation_registry"]
entry = registry.resolve(1)
assert entry is not None
assert entry.locator["run_id"] == "run_abc-123"
message = result.update["messages"][0]
assert "[1]" in message.content
assert "run_abc-123" in message.content
async def test_runtime_survives_langchain_arg_parsing(isolate):
"""runtime must survive langchain arg parsing (else ToolNode drops it)."""
cap = _capability(name="web.scrape", output=_EchoOutput(echoed="hi"))
tools = isolate.module.build_capability_tools(workspace_id=7, capabilities=[cap])
tool = _verb_tool(tools, "web_scrape")
parsed = tool._parse_input({"text": "x", "runtime": "RT"}, "tc_1")
assert parsed["runtime"] == "RT"
async def test_tool_charges_owner(isolate):
output = _EchoOutput(echoed="hi")
cap = _capability(name="web.scrape", output=output)
tools = isolate.module.build_capability_tools(workspace_id=7, capabilities=[cap])
tool = _verb_tool(tools, "web_scrape")
await tool.ainvoke({"text": "ping"})
await _invoke(tool, "ping")
isolate.charge.assert_awaited_once()
(charged_output, unit, ctx), _ = isolate.charge.call_args
@ -136,7 +173,7 @@ async def test_over_budget_returns_friendly_message(isolate):
tools = isolate.module.build_capability_tools(workspace_id=7, capabilities=[cap])
tool = _verb_tool(tools, "web_scrape")
result = await tool.ainvoke({"text": "ping"})
result = await _invoke(tool, "ping")
assert isinstance(result, str)
assert "credit" in result.lower()

View file

@ -0,0 +1,43 @@
"""Unit tests for registering a scraper run as a citation."""
from __future__ import annotations
import pytest
from app.agents.chat.multi_agent_chat.shared.citations import (
CitationRegistry,
CitationSourceType,
to_frontend_payload,
)
from app.capabilities.core.access.run_citation import attach_run_citation
pytestmark = pytest.mark.unit
def test_attaches_run_and_returns_label_with_ordinal() -> None:
registry = CitationRegistry()
n, label = attach_run_citation(
registry, run_external_id="run_abc-123", capability="walmart.scrape"
)
assert n == 1
assert f"[{n}]" in label
entry = registry.resolve(n)
assert entry is not None
assert entry.source_type is CitationSourceType.RUN
assert to_frontend_payload(entry) == "run_abc-123"
def test_same_run_dedups_to_one_label() -> None:
registry = CitationRegistry()
first, _ = attach_run_citation(
registry, run_external_id="run_x", capability="walmart.scrape"
)
again, _ = attach_run_citation(
registry, run_external_id="run_x", capability="walmart.reviews"
)
assert first == again
assert len(registry.by_n) == 1