Repository navigation
Conversation
Inbound quotes already work: the adapter reads message_reference off incoming messages and builds a Reply component from it. The outbound half was never wired up, so a plugin quoting a message with Comp.Reply(id=...) produced plain text with no quote bubble. _parse_to_qqofficial() now returns the first Reply component's message id alongside the media fields, and both send paths (_post_send_one and _send_by_session_common) attach it as message_reference. All four send APIs used here (v2 groups, v2 C2C, guild channel, guild DM) accept that field. The active-send fallback drops the reference together with msg_id, so a stale quote id degrades to plain text instead of blocking the message.
…nd fallback Address review findings: - The passive-to-active fallback stripped message_reference together with msg_id on any failure, so an unrelated send error (markdown rejection, transient server error) lost a perfectly valid quote context. The active retry now keeps the reference first and only drops it when the retry with the reference also fails, since a dead reference id must not block the message body either. - The Reply extraction tests only covered Reply at the chain start. Added a Plain-before-Reply parser case and two fallback-behavior cases.
When the QQ official API refuses a native markdown payload, the fallback used to drop msg_id and resend through the active-message interface, which runs into the 304049 active-message rate limit and loses the reply. Reorder the fallback so a rejected markdown degrades to a plain content reply that keeps msg_id (passive reply, msg_type reset from 2 to 0, stream trailing newline restored). The stream newline repair retry also keeps msg_id. Only errors that are not markdown rejections take the active-send fallback, which keeps the message_reference two-stage retry from the previous fix. Fallback raises are chained so the original rejection stays visible in logs.
When a plugin tool returns ImageContent and the provider does not explicitly declare image support, the cached image was still appended to the context as an ImageURLPart, causing a 400 that aborted the entire turn. Now only append cached images when the provider config explicitly lists 'image' in modalities. The text description in the tool result already tells the LLM the image exists and where it is cached. Fixes AstrBotDevs#10702
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Hey - I've found 5 issues
Prompt for AI Agents
Please address the comments from this code review:
## Individual Comments
### Comment 1
<location path="astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py" line_range="660" />
<code_context>
+ fallback_content = cast(str, fallback_payload.get("content") or "")
+ if fallback_content and not fallback_content.endswith("\n"):
+ fallback_payload["content"] = fallback_content + "\n"
+ return await send_func(fallback_payload)
# QQ 流式 markdown 分片校验：内容必须以换行结尾。
</code_context>
<issue_to_address>
**QQ replies are lost**
When QQ rejects a Markdown send and the plain-content retry also fails with an error recoverable by active sending, `_send_with_markdown_fallback` lets the plain-content retry’s exception escape, so it never reaches the active-send fallback and the reply is not delivered.
Catch the plain-content retry error and continue to the active-send fallback.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:689`.
</issue_to_address>
### Comment 2
<location path="astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py" line_range="689" />
<code_context>
return await send_func(payload)
except _QQOFFICIAL_SEND_API_ERRORS as err:
- logger.info("[QQOfficial] 回复消息失败: %s, 尝试使用主动发送接口。", err)
- if payload.get("msg_id"):
+ # markdown 被服务端拒绝（50056）时，保留 msg_id 退化为纯 content
+ # 被动回复。先去掉 msg_id 会把被动回复变成主动消息，既烧主动
</code_context>
<issue_to_address>
**Quoted friend replies are lost**
When a friend-session send has no cached `msg_id` and QQ rejects its `message_reference`, `_send_by_session_common` omits `msg_id` for friend sends, while `_send_with_markdown_fallback` only retries without `message_reference` inside the `msg_id` branch; the QQ error escapes and the reply body is lost.
Allow the rejected-reference retry independently of whether the payload has a `msg_id`.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_platform_adapter.py:414`.
</issue_to_address>
### Comment 3
<location path="astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py" line_range="703" />
<code_context>
+ raise fallback_err from err
+ # 主动发送仍失败，引用 id 可能已失效；去掉引用最后重试一次，
+ # 失效引用不应阻断正文送达
+ retry_payload = fallback_payload.copy()
+ retry_payload.pop("message_reference", None)
+ try:
</code_context>
<issue_to_address>
**Valid reply quotes disappear**
When an active send with a valid reference gets an unrelated API error and the retry without the reference succeeds, `_send_with_markdown_fallback` removes `message_reference` after any active-send API error, so a successful retry delivers an unquoted reply; when QQ accepted the referenced send but returned a server error, the retry can also deliver a duplicate.
Retry without the reference only when the API error identifies a rejected reference, and do not retry ambiguous server errors.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:704-709`.
</issue_to_address>
### Comment 4
<location path="astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py" line_range="458-459" />
<code_context>
payload["msg_seq"] = random.randint(1, 10000)
+ # 插件通过 Reply 组件给出的引用 id，四个发送接口（群/C2C/频道/频道私信）
+ # 都以 message_reference 接受；拿不到 id 时保持原样，不影响正文发送
+ if reply_message_id:
+ payload["message_reference"] = {"message_id": reply_message_id}
+
</code_context>
<issue_to_address>
**New QQ comments are Chinese**
When the newly added QQ Official comments and fallback log are reviewed or emitted, the added comments and quote-fallback log remain in Chinese, so maintainers receive non-English implementation guidance and diagnostics contrary to the repository convention.
Translate the newly added comments and log message into English.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:639-641`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:664`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:687`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:692-693`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:701-702`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:707`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:995-996`.
</issue_to_address>
### Comment 5
<location path="astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py" line_range="994-999" />
<code_context>
file_source = file_path
elif i.url:
file_source = i.url
+ elif isinstance(i, Reply) and not reply_message_id:
+ # 出站引用回复：取第一个带被引用消息 id 的 Reply 组件
+ quoted_id = str(i.id).strip()
+ if quoted_id:
+ reply_message_id = quoted_id
else:
logger.debug(f"qq_official 忽略 {i.type}")
return (
</code_context>
<issue_to_address>
**Unrelated QQ quoting broadens scope**
When the PR's stated problem remains limited to tool-result images for text-only providers, the new QQ Official outbound quote handling adds behavior unrelated to the text-only provider image failure, broadening this PR beyond the problem it claims to fix.
Move the QQ outbound quote-support changes into a separate PR unless they are explicitly part of this issue's scope.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:459-460`, `astrbot/core/platform/sources/qqofficial/qqofficial_platform_adapter.py:413-414`.
</issue_to_address>Sourcery assessment
Needs a human reviewer. 3 findings to address first, and if QQ reports an error after accepting a send, the retries can deliver a duplicate; the final retry can also send without the requested quote. Delivered messages cannot be recalled by reverting this change.
Blocking findings: astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:660, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:689, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:703
Comment on lines
+458
to
+459
| # 都以 message_reference 接受；拿不到 id 时保持原样，不影响正文发送 | ||
| if reply_message_id: |
Contributor
There was a problem hiding this comment.
⚪ Low · New QQ comments are Chinese
When the newly added QQ Official comments and fallback log are reviewed or emitted, the added comments and quote-fallback log remain in Chinese, so maintainers receive non-English implementation guidance and diagnostics contrary to the repository convention.
Translate the newly added comments and log message into English.
Also at astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:639-641, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:664, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:687, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:692-693, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:701-702, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:707, astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:995-996.
Prompt for AI agents
In `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py` at lines 458-459:
**New QQ comments are Chinese**
When the newly added QQ Official comments and fallback log are reviewed or emitted, the added comments and quote-fallback log remain in Chinese, so maintainers receive non-English implementation guidance and diagnostics contrary to the repository convention.
Translate the newly added comments and log message into English.
Also at `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:639-641`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:664`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:687`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:692-693`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:701-702`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:707`, `astrbot/core/platform/sources/qqofficial/qqofficial_message_event.py:995-996`.From AGENTS.md L50:
5. Use **English** for all comments and logs.Cover empty, null, text-only, and explicitly image-capable providers through the real tool-loop runner. Providers without explicit image support keep the cached-image text without loading or appending image data.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #10702
Problem
When a plugin tool returns
ImageContentand the current provider does not support multimodal input, the entire turn can fail with a 400 response such as "Multimodal data provided, but model does not support multimodal requests." The provider error does not identify the tool result that introduced the image.Root cause
tool_loop_agent_runner.pycaches tool-result images and appends them to the context asImageURLPartuser messages. Itssupports_imagecheck treated an empty or missingmodalitiesvalue as image support for backward compatibility, so providers with unknown or text-only capabilities could receive image data.Tool-result images differ from initial request context: they introduce new multimodal content that the provider did not supply or request.
Fix
"image"inmodalities.modalitiesvalues as unknown capabilities. Skip the image append and avoid reading the image back as base64.tool_loop_agent_runner.pyand its runner regression test. The unrelated QQ Official commits have been reverted from this PR.Testing
ToolLoopAgentRunnerwith a mixedImageContentandTextContenttool result. Against the old upstream implementation, the empty-capabilities case appended an unwantedImageURLPart; the fixed branch skips it. Explicit null, text-only withtool_use, and image-capable withtool_useconfigurations are covered as well.python -I -m pytest tests/test_tool_loop_agent_runner.py -q: 64 passed.tests/unit/test_astr_main_agent.py: 12 passed.docs/tests: 4877 passed, 14 skipped, 2 failed. Both failures are the twotest_config_update_revokes_only_affected_shell_sessionscases, and both reproduce unchanged on upstream base832fc6d98.git diff --checkpassed on the changed files.Checklist
AI Assistance Disclosure
Claude Code assisted with investigating the issue, implementing the fix, and running the regression checks.