Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
#
|
|
|
|
|
# Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
|
|
|
|
|
#
|
|
|
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
|
# you may not use this file except in compliance with the License.
|
|
|
|
|
#
|
|
|
|
|
# You may obtain a copy of the License at
|
|
|
|
|
#
|
|
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
|
#
|
|
|
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
|
|
|
# distributed under the License on an "AS IS" BASIS, WITHOUT WARRANTIES
|
|
|
|
|
# OR CONDITIONS OF ANY KIND, either express or implied. See the License
|
|
|
|
|
# for the specific language governing permissions and limitations under
|
|
|
|
|
# the License.
|
|
|
|
|
#
|
|
|
|
|
|
|
|
|
|
"""SSRF-guard regression tests for rag.app.naive.Markdown.load_images_from_urls.
|
|
|
|
|
|
|
|
|
|
Image references are parsed out of the (untrusted) uploaded markdown document
|
|
|
|
|
and fetched server-side, so the loader must validate + DNS-pin every hop before
|
|
|
|
|
connecting. These tests assert that internal/loopback targets are rejected,
|
|
|
|
|
that redirects to internal targets are rejected, and that a legitimate public
|
|
|
|
|
image is still fetched.
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
from __future__ import annotations
|
|
|
|
|
|
|
|
|
|
import io
|
|
|
|
|
import sys
|
Fix: UserFillUp interactive forms not working in agent explore mode (#14589)
## Summary
- **Backend**: `_iter_session_completion_events` in `agent_api.py` was
filtering out `user_inputs` and `workflow_finished` SSE events, causing
agents with UserFillUp components to silently fail in explore mode — the
interactive form never appeared, while the same agent worked correctly
in run (editor) mode.
- **Frontend**: `SessionChat` component in explore mode was missing
`DebugContent` children rendering inside `MessageItem`, so even if the
backend forwarded the events, the form UI would not render. Added
`DebugContent`, `MarkdownContent`, `useAwaitCompentData` hook, and
input-disabling logic to match the run mode's `chat/box.tsx` behavior.
## What was changed
### Backend (`api/apps/restful_apis/agent_api.py`)
- Line 266: Added `"user_inputs"` and `"workflow_finished"` to the
allowed event filter in `_iter_session_completion_events`
### Frontend (`web/src/pages/agent/explore/components/session-chat.tsx`)
- Added imports: `DebugContent`, `MarkdownContent`,
`useAwaitCompentData`, `useParams`
- Added `sendFormMessage` from `useSendSessionMessage()` hook
- Added `useAwaitCompentData` hook for form state management
- Added `DebugContent` as `MessageItem` children for the latest
assistant message (renders UserFillUp form)
- Added `MarkdownContent` + submitted values display for previous
assistant messages
- Updated `NextMessageInput` disabled states to respect `isWaitting`
(form submission in progress)
## Test plan
- [x] Agent with UserFillUp component (e.g., email draft with
send/edit/cancel options) shows interactive form in **explore mode**
- [x] Same agent continues to work correctly in **run (editor) mode**
- [x] Form submission sends data back to the agent and workflow
continues
- [x] Input field is disabled while waiting for form submission
- [ ] Agents without UserFillUp components are unaffected in explore
mode
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
2026-06-28 21:57:57 +08:00
|
|
|
from importlib import import_module, reload
|
Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
from unittest.mock import MagicMock, patch
|
|
|
|
|
|
|
|
|
|
import pytest
|
|
|
|
|
from PIL import Image
|
|
|
|
|
|
|
|
|
|
from common import ssrf_guard
|
Fix: UserFillUp interactive forms not working in agent explore mode (#14589)
## Summary
- **Backend**: `_iter_session_completion_events` in `agent_api.py` was
filtering out `user_inputs` and `workflow_finished` SSE events, causing
agents with UserFillUp components to silently fail in explore mode — the
interactive form never appeared, while the same agent worked correctly
in run (editor) mode.
- **Frontend**: `SessionChat` component in explore mode was missing
`DebugContent` children rendering inside `MessageItem`, so even if the
backend forwarded the events, the form UI would not render. Added
`DebugContent`, `MarkdownContent`, `useAwaitCompentData` hook, and
input-disabling logic to match the run mode's `chat/box.tsx` behavior.
## What was changed
### Backend (`api/apps/restful_apis/agent_api.py`)
- Line 266: Added `"user_inputs"` and `"workflow_finished"` to the
allowed event filter in `_iter_session_completion_events`
### Frontend (`web/src/pages/agent/explore/components/session-chat.tsx`)
- Added imports: `DebugContent`, `MarkdownContent`,
`useAwaitCompentData`, `useParams`
- Added `sendFormMessage` from `useSendSessionMessage()` hook
- Added `useAwaitCompentData` hook for form state management
- Added `DebugContent` as `MessageItem` children for the latest
assistant message (renders UserFillUp form)
- Added `MarkdownContent` + submitted values display for previous
assistant messages
- Updated `NextMessageInput` disabled states to respect `isWaitting`
(form submission in progress)
## Test plan
- [x] Agent with UserFillUp component (e.g., email draft with
send/edit/cancel options) shows interactive form in **explore mode**
- [x] Same agent continues to work correctly in **run (editor) mode**
- [x] Form submission sends data back to the agent and workflow
continues
- [x] Input field is disabled while waiting for form submission
- [ ] Agents without UserFillUp components are unaffected in explore
mode
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
2026-06-28 21:57:57 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
|
|
|
def naive_module():
|
|
|
|
|
"""Load rag.app.naive with heavy optional dependencies stubbed locally."""
|
|
|
|
|
stub_names = [
|
|
|
|
|
"deepdoc.vision.ocr",
|
|
|
|
|
"deepdoc.parser.figure_parser",
|
|
|
|
|
"deepdoc.parser.docling_parser",
|
|
|
|
|
"deepdoc.parser.tcadp_parser",
|
|
|
|
|
"rag.app.picture",
|
|
|
|
|
]
|
|
|
|
|
original_modules = {name: sys.modules.get(name) for name in stub_names}
|
|
|
|
|
|
|
|
|
|
try:
|
|
|
|
|
for name in stub_names:
|
|
|
|
|
sys.modules[name] = MagicMock()
|
|
|
|
|
module = import_module("rag.app.naive")
|
|
|
|
|
module = reload(module)
|
|
|
|
|
yield module
|
|
|
|
|
finally:
|
|
|
|
|
for name, original in original_modules.items():
|
|
|
|
|
if original is None:
|
|
|
|
|
sys.modules.pop(name, None)
|
|
|
|
|
else:
|
|
|
|
|
sys.modules[name] = original
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
|
|
|
def max_image_redirects(naive_module):
|
|
|
|
|
return naive_module.MAX_IMAGE_REDIRECTS
|
Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def _png_bytes() -> bytes:
|
|
|
|
|
buf = io.BytesIO()
|
|
|
|
|
Image.new("RGB", (1, 1), (255, 0, 0)).save(buf, format="PNG")
|
|
|
|
|
return buf.getvalue()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class _Resp:
|
|
|
|
|
"""Minimal stand-in for requests.Response."""
|
|
|
|
|
|
|
|
|
|
def __init__(self, status_code, headers=None, content=b""):
|
|
|
|
|
self.status_code = status_code
|
|
|
|
|
self.headers = headers or {}
|
|
|
|
|
self.content = content
|
|
|
|
|
|
|
|
|
|
def close(self):
|
|
|
|
|
pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.fixture
|
Fix: UserFillUp interactive forms not working in agent explore mode (#14589)
## Summary
- **Backend**: `_iter_session_completion_events` in `agent_api.py` was
filtering out `user_inputs` and `workflow_finished` SSE events, causing
agents with UserFillUp components to silently fail in explore mode — the
interactive form never appeared, while the same agent worked correctly
in run (editor) mode.
- **Frontend**: `SessionChat` component in explore mode was missing
`DebugContent` children rendering inside `MessageItem`, so even if the
backend forwarded the events, the form UI would not render. Added
`DebugContent`, `MarkdownContent`, `useAwaitCompentData` hook, and
input-disabling logic to match the run mode's `chat/box.tsx` behavior.
## What was changed
### Backend (`api/apps/restful_apis/agent_api.py`)
- Line 266: Added `"user_inputs"` and `"workflow_finished"` to the
allowed event filter in `_iter_session_completion_events`
### Frontend (`web/src/pages/agent/explore/components/session-chat.tsx`)
- Added imports: `DebugContent`, `MarkdownContent`,
`useAwaitCompentData`, `useParams`
- Added `sendFormMessage` from `useSendSessionMessage()` hook
- Added `useAwaitCompentData` hook for form state management
- Added `DebugContent` as `MessageItem` children for the latest
assistant message (renders UserFillUp form)
- Added `MarkdownContent` + submitted values display for previous
assistant messages
- Updated `NextMessageInput` disabled states to respect `isWaitting`
(form submission in progress)
## Test plan
- [x] Agent with UserFillUp component (e.g., email draft with
send/edit/cancel options) shows interactive form in **explore mode**
- [x] Same agent continues to work correctly in **run (editor) mode**
- [x] Form submission sends data back to the agent and workflow
continues
- [x] Input field is disabled while waiting for form submission
- [ ] Agents without UserFillUp components are unaffected in explore
mode
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
2026-06-28 21:57:57 +08:00
|
|
|
def parser(naive_module):
|
|
|
|
|
return naive_module.Markdown(128)
|
Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.p1
|
|
|
|
|
def test_blocks_internal_url_without_fetching(parser):
|
|
|
|
|
"""A markdown image pointing at an internal host must never be requested."""
|
|
|
|
|
with (
|
|
|
|
|
patch.object(ssrf_guard, "assert_url_is_safe", side_effect=ValueError("non-public")) as guard,
|
|
|
|
|
patch("requests.get") as get,
|
|
|
|
|
):
|
|
|
|
|
images, cache = parser.load_images_from_urls(["http://169.254.169.254/latest/meta-data/"])
|
|
|
|
|
|
|
|
|
|
guard.assert_called_once()
|
|
|
|
|
get.assert_not_called() # SSRF guard rejects before any connection is made
|
|
|
|
|
assert images == []
|
|
|
|
|
assert cache["http://169.254.169.254/latest/meta-data/"] is None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.p1
|
|
|
|
|
def test_blocks_redirect_to_internal_target(parser):
|
|
|
|
|
"""A public URL that 302-redirects to a loopback target must be rejected."""
|
|
|
|
|
|
|
|
|
|
def selective_assert(url, **kwargs):
|
|
|
|
|
if "127.0.0.1" in url or "localhost" in url:
|
|
|
|
|
raise ValueError("redirect resolves to non-public address")
|
|
|
|
|
return ("public.example", "8.8.8.8")
|
|
|
|
|
|
|
|
|
|
redirect = _Resp(302, headers={"Location": "http://127.0.0.1/secret"})
|
|
|
|
|
with (
|
|
|
|
|
patch.object(ssrf_guard, "assert_url_is_safe", side_effect=selective_assert),
|
|
|
|
|
patch("requests.get", return_value=redirect) as get,
|
|
|
|
|
):
|
|
|
|
|
images, _ = parser.load_images_from_urls(["http://public.example/logo.png"])
|
|
|
|
|
|
|
|
|
|
# Only the first (public) hop is fetched; the redirect target is blocked
|
|
|
|
|
# by re-validation before a second request is made.
|
|
|
|
|
assert get.call_count == 1
|
|
|
|
|
assert images == []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.p1
|
|
|
|
|
def test_fetches_legitimate_public_image(parser):
|
|
|
|
|
png = _png_bytes()
|
|
|
|
|
ok = _Resp(200, headers={"Content-Type": "image/png"}, content=png)
|
|
|
|
|
with (
|
|
|
|
|
patch.object(ssrf_guard, "assert_url_is_safe", return_value=("public.example", "8.8.8.8")),
|
|
|
|
|
patch("requests.get", return_value=ok) as get,
|
|
|
|
|
):
|
|
|
|
|
images, cache = parser.load_images_from_urls(["http://public.example/logo.png"])
|
|
|
|
|
|
|
|
|
|
get.assert_called_once()
|
|
|
|
|
# allow_redirects must be disabled so redirects are validated per hop.
|
|
|
|
|
assert get.call_args.kwargs.get("allow_redirects") is False
|
|
|
|
|
assert len(images) == 1
|
|
|
|
|
assert isinstance(images[0], Image.Image)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@pytest.mark.p1
|
Fix: UserFillUp interactive forms not working in agent explore mode (#14589)
## Summary
- **Backend**: `_iter_session_completion_events` in `agent_api.py` was
filtering out `user_inputs` and `workflow_finished` SSE events, causing
agents with UserFillUp components to silently fail in explore mode — the
interactive form never appeared, while the same agent worked correctly
in run (editor) mode.
- **Frontend**: `SessionChat` component in explore mode was missing
`DebugContent` children rendering inside `MessageItem`, so even if the
backend forwarded the events, the form UI would not render. Added
`DebugContent`, `MarkdownContent`, `useAwaitCompentData` hook, and
input-disabling logic to match the run mode's `chat/box.tsx` behavior.
## What was changed
### Backend (`api/apps/restful_apis/agent_api.py`)
- Line 266: Added `"user_inputs"` and `"workflow_finished"` to the
allowed event filter in `_iter_session_completion_events`
### Frontend (`web/src/pages/agent/explore/components/session-chat.tsx`)
- Added imports: `DebugContent`, `MarkdownContent`,
`useAwaitCompentData`, `useParams`
- Added `sendFormMessage` from `useSendSessionMessage()` hook
- Added `useAwaitCompentData` hook for form state management
- Added `DebugContent` as `MessageItem` children for the latest
assistant message (renders UserFillUp form)
- Added `MarkdownContent` + submitted values display for previous
assistant messages
- Updated `NextMessageInput` disabled states to respect `isWaitting`
(form submission in progress)
## Test plan
- [x] Agent with UserFillUp component (e.g., email draft with
send/edit/cancel options) shows interactive form in **explore mode**
- [x] Same agent continues to work correctly in **run (editor) mode**
- [x] Form submission sends data back to the agent and workflow
continues
- [x] Input field is disabled while waiting for form submission
- [ ] Agents without UserFillUp components are unaffected in explore
mode
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
2026-06-28 21:57:57 +08:00
|
|
|
def test_redirect_chain_is_bounded(parser, max_image_redirects):
|
Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
"""An endless redirect loop is abandoned instead of being followed forever."""
|
|
|
|
|
loop = _Resp(302, headers={"Location": "http://public.example/next"})
|
|
|
|
|
with (
|
|
|
|
|
patch.object(ssrf_guard, "assert_url_is_safe", return_value=("public.example", "8.8.8.8")),
|
|
|
|
|
patch("requests.get", return_value=loop) as get,
|
|
|
|
|
):
|
|
|
|
|
images, _ = parser.load_images_from_urls(["http://public.example/start"])
|
|
|
|
|
|
Fix: UserFillUp interactive forms not working in agent explore mode (#14589)
## Summary
- **Backend**: `_iter_session_completion_events` in `agent_api.py` was
filtering out `user_inputs` and `workflow_finished` SSE events, causing
agents with UserFillUp components to silently fail in explore mode — the
interactive form never appeared, while the same agent worked correctly
in run (editor) mode.
- **Frontend**: `SessionChat` component in explore mode was missing
`DebugContent` children rendering inside `MessageItem`, so even if the
backend forwarded the events, the form UI would not render. Added
`DebugContent`, `MarkdownContent`, `useAwaitCompentData` hook, and
input-disabling logic to match the run mode's `chat/box.tsx` behavior.
## What was changed
### Backend (`api/apps/restful_apis/agent_api.py`)
- Line 266: Added `"user_inputs"` and `"workflow_finished"` to the
allowed event filter in `_iter_session_completion_events`
### Frontend (`web/src/pages/agent/explore/components/session-chat.tsx`)
- Added imports: `DebugContent`, `MarkdownContent`,
`useAwaitCompentData`, `useParams`
- Added `sendFormMessage` from `useSendSessionMessage()` hook
- Added `useAwaitCompentData` hook for form state management
- Added `DebugContent` as `MessageItem` children for the latest
assistant message (renders UserFillUp form)
- Added `MarkdownContent` + submitted values display for previous
assistant messages
- Updated `NextMessageInput` disabled states to respect `isWaitting`
(form submission in progress)
## Test plan
- [x] Agent with UserFillUp component (e.g., email draft with
send/edit/cancel options) shows interactive form in **explore mode**
- [x] Same agent continues to work correctly in **run (editor) mode**
- [x] Form submission sends data back to the agent and workflow
continues
- [x] Input field is disabled while waiting for form submission
- [ ] Agents without UserFillUp components are unaffected in explore
mode
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
2026-06-28 21:57:57 +08:00
|
|
|
assert get.call_count == max_image_redirects + 1
|
Fix: SSRF in markdown parser remote image fetch (#15438)
### What problem does this PR solve?
`rag/app/naive.py` `Markdown.load_images_from_urls` fetched image URLs
parsed
straight out of an untrusted uploaded markdown document via a raw
`requests.get`,
with no SSRF validation. Markdown chunking always reaches this path
(`return_section_images=True`), so any authenticated user who uploads a
`.md`/`.markdown`/`.mdx` file to a knowledge base could make the server
issue
requests to internal services or cloud-metadata endpoints, e.g.
``. The `image/`
Content-Type
check only gates decoding — the outbound request (the SSRF) always
fires.
This was the one user-controlled fetch site missed by the project's
existing
SSRF-hardening (`common/ssrf_guard.py`, already applied to the crawler,
SearXNG,
RSS connector, MCP/document APIs, and OAuth avatar download).
The fix validates and DNS-pins every hop with
`common.ssrf_guard.assert_url_is_safe`
before connecting, and follows redirects manually so each redirect
target is
re-validated (closing the DNS-rebinding / redirect-bypass window),
mirroring
`common/data_source/rss_connector.py`. Blocked URLs are skipped and
logged like
any other unreachable image, so legitimate public images are unaffected.
Adds a
regression test at `test/unit_test/rag/app/test_markdown_image_ssrf.py`.
Closes #15437
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Ubuntu <ubuntu@ubuntu-2204.linuxvmimages.local>
Co-authored-by: galuis116 <galuis116@users.noreply.github.com>
2026-06-16 03:54:55 -07:00
|
|
|
assert images == []
|