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
- 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
* feat(extension): add PluginSortControl reusable component for sorting
* i18n: add i18n keys for plugin sorting and filtering features
* feat(extension): add sorting and status filtering for installed plugins
Backend changes (plugin.py):
- Add _resolve_plugin_dir method to resolve plugin directory path
- Add _get_plugin_installed_at method to get installation time from file mtime
- Add installed_at field to plugin API response
Frontend changes (InstalledPluginsTab.vue):
- Import PluginSortControl component
- Add status filter toggle (all/enabled/disabled) using v-btn-toggle
- Integrate PluginSortControl for sorting options
- Add toolbar layout with actions and controls sections
Frontend changes (MarketPluginsTab.vue):
- Import PluginSortControl component
- Replace v-select + v-btn combination with unified PluginSortControl
Frontend changes (useExtensionPage.js):
- Add installedStatusFilter, installedSortBy, installedSortOrder refs
- Add installedSortItems and installedSortUsesOrder computed properties
- Add sortInstalledPlugins function with multi-criteria support
- Support sorting by install time, name, author, and update status
- Add status filtering in filteredPlugins computed property
- Disable default table sorting by setting sortable: false
* test: add tests for installed_at field in plugin API
- Assert all plugins have installed_at field in get_plugins response
- Assert installed_at is not null after plugin installation
* fix(extension): add explicit fallbacks for installed plugin sort comparisons
* i18n(extension): rename install time label to last modified
* fix(extension): cache installed_at parsing and validate timestamp format in tests
* test(dashboard): strengthen installed_at coverage for plugin API
* fix(provider): handle MiniMax ThinkingBlock when max_tokens reached
Fixes#5912
Problem: MiniMax API returns ThinkingBlock when stop_reason='max_tokens',
but AstrBot throws 'completion 无法解析' exception because both
completion_text and tools_call_args are empty.
Root cause: The validation logic didn't consider ThinkingBlock
(reasoning_content) as valid content.
Fix: When completion_text and tools_call_args are empty but
reasoning_content is present, treat it as valid instead of throwing
exception. This happens when the model thinks but runs out of tokens
before generating the actual response.
Impact: MiniMax models now work correctly when responses are truncated
due to max_tokens limit.
* refactor: address review feedback
1. Use getattr for safe stop_reason access (prevent AttributeError)
2. Use ValueError instead of generic Exception for better error handling
Thanks @gemini-code-assist and @sourcery-ai for the review!
* refactor: flatten nested if/else with guard clause
Address Gemini Code Assist feedback:
- Use guard clause for early return
- Flattened nested conditional for better readability
Logic unchanged, just cleaner code structure.
* fix(provider): improve logging for ThinkingBlock completions in ProviderAnthropic
---------
Co-authored-by: Soulter <905617992@qq.com>
* fix: prevent crash on malformed MCP server config (#5666)
* fix: prevent crash on malformed MCP server config (#5666)
* fix: validate MCP connection before persisting server config
* fix: guard mcpServers type before iterating server list
* refactor: use typed empty-config error and extract MCP rollback helper
* fix: translate error messages and comments to English for consistency
---------
Co-authored-by: Soulter <905617992@qq.com>
* fix(extension): support searching installed plugins by display name
* fix: unify plugin search matching across installed and market tabs
* refactor(extension): optimize plugin search matcher and remove redundant checks
* refactor(extension-page): centralize search query normalization and text matching logic
- Extract `buildSearchQuery` to create normalized query objects from raw input
- Extract `matchesText` as a reusable text matching helper for normalized/loose/pinyin/initials matching
- Remove unused `marketCustomFilter` to eliminate dead code
- Simplify `matchesPluginSearch` to accept query object instead of pre-normalized string
- Replace Set with Array for candidates to simplify control flow
- Avoid redundant normalization by having callers pass raw strings to `buildSearchQuery`
* refactor: remove unused marketCustomFilter from extension page components
- Remove marketCustomFilter from destructuring in ExtensionPage.vue, InstalledPluginsTab.vue, and MarketPluginsTab.vue
* refactor(extension): extract plugin search utilities into shared module
- Create pluginSearch.js to centralize plugin search helpers
- Move `normalizeStr`, `normalizeLoose`, `toPinyinText`, and `toInitials` into the shared module
- Add `buildSearchQuery`, `matchesText`, and `matchesPluginSearch` for reusable search matching
- Refactor useExtensionPage.js to consume the shared utilities
- Simplify plugin search logic by consolidating normalization and matching in one place
* refactor(extension): add caching to pinyin utilities and extract search fields helper
- Add Map-based caching for `toPinyinText` and `toInitials` to avoid redundant pinyin computation
- Extract `getPluginSearchFields` function to retrieve plugin fields for searching
- Improve plugin search performance with caching and better code organization
* perf(extension): add bounded caching for plugin search
- cap normalization and pinyin caches with `MAX_SEARCH_CACHE_SIZE`
- add `setCacheValue()` for oldest-entry eviction
- cache normalized and loose text values to avoid repeated string processing
- skip pinyin matching for non-CJK text using Unicode `\p{Unified_Ideograph}` property
- improve search performance while keeping memory usage bounded
* refactor(extension): extract memoizeLRU helper for cache management
- Create `memoizeLRU` higher-order function to generate LRU-cached functions
- Replace manual cache implementation with `memoizeLRU` for cleaner code
- Optimize `matchesText` to lazily compute looseValue only when needed
- Simplify caching logic while maintaining bounded cache size
* refactor(extension): simplify memoization and remove LRU logic
- Rename `memoizeLRU` to `memoizeStringFn` and remove bounded cache size
- Simplify cache hit logic for cleaner code
- Remove `MAX_SEARCH_CACHE_SIZE` constant as it's no longer needed
* fix: apply reply_with_quote and reply_with_mention to image-only responses
* fix: restrict reply_with_quote and reply_with_mention to plain-text/image chains
* feat(skills): add batch upload functionality for multiple skill ZIP files
- Implemented a new endpoint for batch uploading skills.
- Enhanced the SkillsSection component to support multiple file selection and drag-and-drop functionality.
- Updated localization files for new upload features and messages.
- Added tests to validate batch upload behavior and error handling.
* feat(skills): improve batch upload handling and enhance accessibility for dropzone
* feat(skills): enhance batch upload process and improve UI for better user experience
* feat(skills): enhance skills upload dialog layout and styling for improved usability
* feat(skills): update upload dialog description styling for better visibility and usability
* feat(skills): improve upload dialog button styling and layout for enhanced usability
* feat(skills): refine upload dialog text for clarity and consistency
* feat(skills): enhance batch upload functionality by ignoring __MACOSX entries and improving upload dialog styling
* feat(skills): refactor upload dialog and button styles for improved consistency and usability
---------
Co-authored-by: whatevertogo <whatevertogo@users.noreply.github.com>
* test(skill_manager): update sandbox cache path expectations
adjust sandbox cache tests to match absolute path resolution in
list_skills for sandbox runtime.
verify sandbox-cached skills cannot be deactivated via set_skill_active
by asserting a PermissionError, and keep active-only listing behavior
intact.
add coverage for show_sandbox_path=false to ensure local skills still
override cached metadata while sandbox-only skills retain cached paths.
* test(skill_manager): tighten local skill path assertions
zenfun
advent259141
Soulter
ChuwuYo
Soulter
邹永赫
エイカク
camera-2018
一袋米要扛幾樓
lustresixx
dependabot[bot]
Jason
whatevertogo
sanyekana
Windy_cold
時壹
whatevertogo
Ruochen Pan