wehub-resource-sync
b4fbd6fe9f
Deploy Site / deploy-vercel (push) Has been skipped
Deploy Site / deploy-docs (push) Has been skipped
Build Skills Index / build-index (push) Has been skipped
CI / Deny unrelated histories (push) Has been skipped
CI / Detect affected areas (push) Successful in 27m35s
CI / OSV scan (push) Failing after 4s
CI / Build&Test Docker image (push) Successful in 9s
CI / Supply-chain scan (push) Has been skipped
CI / Lint Docker scripts (push) Failing after 5m13s
CI / Check contributors (push) Failing after 12m8s
CI / Docs Site (push) Failing after 12m8s
CI / TypeScript (push) Failing after 12m8s
CI / Python lints (push) Failing after 12m9s
CI / Python tests (push) Failing after 12m9s
CI / Check uv.lock (push) Failing after 23m22s
CI / CI timing report (push) Has been cancelled
Build Skills Index / trigger-deploy (push) Has been cancelled
CI / All required checks pass (push) Has been cancelled
58 lines
2.4 KiB
Python
58 lines
2.4 KiB
Python
"""Tests for the document context note prepended to user turns with attachments.
|
|
|
|
A user who attaches a PDF / DOCX in chat used to see the agent treat it as
|
|
"unreadable" because the context note told the model to "Ask the user what
|
|
they'd like you to do with it" — steering it away from extracting the text it
|
|
is perfectly capable of reading. These tests pin the contract:
|
|
|
|
- text documents: note confirms the (adapter-)inlined content + records path.
|
|
- binary documents (PDF/DOCX/…): note tells the agent to extract the text
|
|
itself and never tells it to punt back to the user.
|
|
"""
|
|
|
|
import importlib
|
|
|
|
import pytest
|
|
|
|
gateway_run = importlib.import_module("gateway.run")
|
|
_build_document_context_note = gateway_run._build_document_context_note
|
|
|
|
|
|
class TestTextDocumentNote:
|
|
@pytest.mark.parametrize("mtype", ["text/plain", "text/markdown", "text/csv"])
|
|
def test_text_note_mentions_included_content_and_path(self, mtype):
|
|
note = _build_document_context_note("notes.txt", "/cache/doc_notes.txt", mtype)
|
|
assert "text document" in note
|
|
assert "notes.txt" in note
|
|
assert "/cache/doc_notes.txt" in note
|
|
assert "included below" in note
|
|
|
|
|
|
class TestBinaryDocumentNote:
|
|
@pytest.mark.parametrize(
|
|
"mtype",
|
|
[
|
|
"application/pdf",
|
|
"application/vnd.openxmlformats-officedocument.wordprocessingml.document",
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
|
|
"application/octet-stream",
|
|
],
|
|
)
|
|
def test_binary_note_guides_extraction(self, mtype):
|
|
note = _build_document_context_note("contract.pdf", "/cache/doc_contract.pdf", mtype)
|
|
# Records the path so the agent can open it.
|
|
assert "/cache/doc_contract.pdf" in note
|
|
# Tells the agent to read it by extracting the text...
|
|
assert "extract" in note.lower()
|
|
# ...and does NOT steer it into punting back to the user (the bug).
|
|
assert "ask the user" not in note.lower()
|
|
assert "paste" in note.lower()
|
|
|
|
def test_binary_note_distinct_from_text_note(self):
|
|
text_note = _build_document_context_note("a.txt", "/c/a.txt", "text/plain")
|
|
pdf_note = _build_document_context_note("a.pdf", "/c/a.pdf", "application/pdf")
|
|
assert text_note != pdf_note
|
|
# The text path claims content is inlined; the binary path must not.
|
|
assert "included below" in text_note
|
|
assert "included below" not in pdf_note
|