1. Fix missing spaces in cron job wake prompt string concatenation.
Python implicit string concatenation produced:
"...scheduled taskProceed..." and "...conversation.After..."
which sent garbled instructions to the LLM agent, causing unreliable
cron job execution.
2. Replace deprecated datetime.utcnow() with
datetime.now(datetime.timezone.utc) in JWT generation.
utcnow() is deprecated since Python 3.12 and returns naive datetime
which can cause incorrect token expiry on non-UTC systems.
Closes#6103Closes#6165
Co-authored-by: easonysliu <easonysliu@tencent.com>
In _handle_api_error(), when a 429 rate-limit is encountered, the code
calls available_api_keys.remove(chosen_key). If the same key was already
removed in a previous retry iteration (e.g. the key rotated back to the
same value), this raises ValueError which crashes the entire LLM request
with an opaque error instead of a proper retry/fallback.
Add a membership check before calling remove() to prevent the crash.
Co-authored-by: easonysliu <easonysliu@tencent.com>
- .gitignore: keep both .serena and .worktrees/ entries
- astr_main_agent_resources.py: keep deletion (refactored to tools/)
- send_message.py: port video message type support from master
- shipyard_neo: browser property now returns None when not initialized
instead of raising RuntimeError, matching ComputerBooter base contract
- computer_tool_provider: remove dead os.environ writes for shipyard
(SHIPYARD_ENDPOINT / SHIPYARD_ACCESS_TOKEN are never read anywhere)
and remove unused os import
- Rewrite TestApplySandboxToolsRefactored to test ComputerToolProvider
directly (_apply_sandbox_tools was removed; old tests permanently skipped)
- Add TestExecutorCapabilityGuard with 5 strict tests for browser
capability rejection at executor level
- Fix typo: "on hehalf you" -> "on behalf of you" in subagent result
- Remove extra blank lines (ruff E303) after _apply_sandbox_tools comment
- Remove dead _build_sync_and_scan_command (no callers after refactor)
* fix(qqofficial): fix streaming message delivery for C2C
* fix(qqofficial): rewrite send_streaming for C2C vs non-C2C split
* fix(qqofficial): add _extract_response_message_id for safe id extraction
* fix(qqofficial): flush stream segment on tool-call break signal
* fix(qqofficial): downgrade rich-media to non-stream send in C2C
* fix(qqofficial): auto-append \n to final stream chunk (state=10)
* fix(qqofficial): propagate stream param to all _send_with_markdown_fallback call sites
* fix(qqofficial): retry on STREAM_MARKDOWN_NEWLINE_ERROR with newline fix
* fix(qqofficial): handle None/non-dict response in post_c2c_message gracefully
* fix(qqofficial): remove msg_id from video/file media payloads in send_by_session
QQ API rejects msg_id on proactive media (video/file, msg_type=7) messages
sent via the tool-call path, returning "请求参数msg_id无效或越权". The
msg_id passive-reply credential is consumed by the first send and cannot be
reused for subsequent media uploads in the same session.
Remove msg_id from the payload after setting msg_type=7 for video and file
sends, for both FRIEND_MESSAGE (C2C) and GROUP_MESSAGE paths.
* fix(qqofficial): replace deprecated get_event_loop() with get_running_loop()
asyncio.get_event_loop() is deprecated since Python 3.10 and raises a
DeprecationWarning (or errors) when called from inside a running coroutine
without a current event loop set on the thread. Replace both call-sites
in the streaming throttle logic with asyncio.get_running_loop(), which is
the correct API to use inside an already-running async context.
Co-Authored-By: Claude Sonnet <noreply@anthropic.com>
---------
Co-authored-by: 2ndelement <2ndelement@users.noreply.github.com>
Co-authored-by: Claude Sonnet <noreply@anthropic.com>
Two changes to make the tool schema sent to the LLM deterministic:
1. ToolSet.normalize() — sort tools by name before serialization.
Called at the end of build_main_agent() after all injection passes.
Eliminates ordering drift from plugin load order, MCP reconnection,
and persona tool list differences.
2. Always inject full sandbox tool set — ComputerToolProvider now
returns get_default_sandbox_tools() unconditionally, regardless of
sandbox boot state. Browser tools are always in the schema even if
the sandbox profile lacks browser capability. The executor rejects
calls to unavailable browser tools with a descriptive error instead
of silently omitting them from the schema.
This eliminates the pre-boot/post-boot tool set jump that caused
prefix cache misses on the second request of a conversation.
- Add get_default_tools/get_tools/get_system_prompt_parts to ComputerBooter base
- Each booter subclass (ShipyardNeo, Shipyard, Boxlite) declares its own tools
- ComputerToolProvider now delegates to booter API via computer_client helpers
- Add unified query API: get_sandbox_tools, get_default_sandbox_tools, etc.
- Extract Neo prompts to dedicated computer/prompts.py module
- Add booter type constants (booters/constants.py)
- Fix subagent tool path to pass sandbox_cfg and session_id
- Fix Sourcery issues: shell injection in send_message, typo in prompts,
internal tools bypass inactivated_llm_tools check
* feat: add video message support and enhance message type descriptions in SendMessageToUserTool
* feat: add error handling for disabled sandbox runtime in get_booter function
- Add audio data validation in MiniMax TTS get_audio() method to detect empty responses
- Validate generated audio file size in TTSProvider.test() to ensure valid output
- Provide detailed error messages guiding users to check group_id configuration
- Auto-cleanup test audio files after validation
- Fixes issue where 0KB audio files would pass TTS detection when group_id is not configured
