feat: implement ModelMessage storage layer with AI SDK response messages #11409
+2,648
−3,206
Conversation
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
…igration Add foundation infrastructure for migrating message storage from Anthropic format to AI SDK ModelMessage format (EXT-646). - RooMessage type = AI SDK ModelMessage + Roo metadata (ts, condense, truncation) - Anthropic-to-RooMessage converter for migrating old conversations on read - Versioned storage (readRooMessages/saveRooMessages) with auto-detection - flattenModelMessagesToStringContent utility for providers needing string content No behavior change - purely additive. Existing code paths untouched. EXT-647 will wire Task.ts to use these new functions.
…s/index.ts and unnecessary re-exports
- Narrow summary from any[] to Array<{ type: string; text: string }> matching EncryptedReasoningItem
- Delete unused converters/index.ts barrel (knip: unused file)
- Remove UserContent, AssistantContent, ToolContent re-exports (importable from 'ai' directly)
- Remove FlattenMessagesOptions re-export from index.ts
Complete migration from Anthropic.MessageParam to RooMessage (ModelMessage + metadata) for internal message storage and the entire provider pipeline. Key changes: - Task.ts: apiConversationHistory is now RooMessage[], stored via readRooMessages/saveRooMessages - Message construction: userMessageContent uses TextPart/ImagePart, pendingToolResults uses ToolResultPart - ApiHandler.createMessage() accepts RooMessage[] instead of Anthropic.Messages.MessageParam[] - All 30 providers updated: messages passed directly to streamText()/generateText() - Removed convertToAiSdkMessages() calls from all AI SDK providers - buildCleanConversationHistory simplified from ~140 lines to ~20 lines - Resume logic rewritten for RooMessage format - Supporting systems updated: condense, messageManager, ClineProvider Old Anthropic-format conversations auto-convert on first open via readRooMessages(). New conversations stored in versioned v2 format. 5536 tests pass, 0 failures. EXT-647
Replace `as any` / `as unknown as` casts with proper RooMessage types across the storage and API pipeline: - Add UserContentPart alias and content-part type guards to rooMessage.ts - Migrate maybeRemoveImageBlocks, formatResponse, processUserContentMentions, validateAndFixToolResultIds, condense system, and convertToOpenAiMessages from Anthropic SDK types to AI SDK RooMessage types - Change initiateTaskLoop/recursivelyMakeClineRequests signatures from Anthropic.Messages.ContentBlockParam[] to UserContentPart[] - Type the assistant message builder in Task.ts, remove double-casts in the API pipeline - Remove unused Anthropic imports from 7 source files - Update ToolResponse type and presentAssistantMessage to use ImagePart - All functions accept both AI SDK (tool-call/tool-result) and legacy Anthropic (tool_use/tool_result) formats for backward compatibility Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add RooRoleMessage type, dual-format legacy block interfaces (LegacyToolUseBlock, LegacyToolResultBlock), union types (AnyToolCallBlock, AnyToolResultBlock), type guards, and accessor helpers to rooMessage.ts. Replace scattered `as any` casts across condense, context-management, Task.ts, validateToolResultIds.ts, and openai-format.ts with these shared typed utilities. Remove leaked Anthropic SDK import from context-management in favor of a provider-agnostic ContentBlockParam type. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- toolResultToText: handle { type, value } object shape returned by
getToolResultContent for AI SDK ToolResultPart.output, preventing
tool result content from being lost during condensation summaries
- export-markdown: add tool-call/tool-result (AI SDK format) handling
alongside legacy tool_use/tool_result, preventing post-migration
conversations from rendering as [Unexpected content type: tool-call]
Addresses review feedback from PR #11386.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Instead of manually reconstructing assistant messages from stream events and injecting custom blocks (thinking, redacted_thinking, thoughtSignature) via side-channel handler methods, grab the fully-formed AssistantModelMessage from the AI SDK's result.response.messages and store it directly. This eliminates validation errors on subsequent turns caused by custom blocks that don't match the AI SDK's ModelMessage schema. - Add ApiStreamResponseMessageChunk type to stream chunk union - Add yieldResponseMessage helper and wire it into consumeAiSdkStream - All 24 AI SDK providers now yield response_message after streaming - Task.ts captures and stores native-format messages directly - addToApiConversationHistory detects native format (providerOptions on content parts) and skips legacy block injection - buildCleanConversationHistory passes native-format messages through - Remove dead side-channel methods: getThoughtSignature, getRedactedThinkingBlocks, getReasoningDetails, getSummary and their backing fields from all providers - Add redacted_thinking conversion in anthropicToRoo.ts for backward compat Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
daniel-lxs
requested review from
cte,
hannesrudolph,
jr and
mrubens
as code owners
dosubot
bot
added
size:XL
Contributor
Re-reviewed after 984857e. Both commits address all previously flagged issues:
Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues. |
![roomote[bot]](https://avatars.githubusercontent.com/in/1546624?s=60&v=4){kind=small}
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
- Remove redundant double cast in bedrock.ts (as ModelMessage[] as ModelMessage[]) - Use isAnyToolCallBlock/getToolCallId in condense orphan filter to handle both AI SDK tool-call and legacy tool_use formats - Fix misleading comment and remove dead conditional in processUserContentMentions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…dling in Task class
…ve saveApiConversationHistory error handling
daniel-lxs
deleted the
feature/ext-647-implement-modelmessage-storage-layer
branch
hannesrudolph
added a commit
that referenced
this pull request
Feb 11, 2026
Move cache breakpoint logic from individual providers to a shared utility called from Task.ts before createMessage(). Messages arrive at providers pre-annotated with providerOptions, and the AI SDK routes the correct options to the active provider automatically. New files: - src/api/transform/prompt-cache.ts: resolveCacheProviderOptions() + applyCacheBreakpoints() with provider adapter mapping - src/api/transform/__tests__/prompt-cache.spec.ts: 14 test cases Changes per provider: - anthropic.ts: removed targeting block + applyCacheControlToAiSdkMessages() - anthropic-vertex.ts: same - minimax.ts: same - bedrock.ts: removed targeting block + applyCachePointsToAiSdkMessages() Key improvements: - Targets non-assistant batches (user + tool) instead of only role=user. After PR #11409, tool results are separate role=tool messages that now correctly receive cache breakpoints. - Single source of truth: cache strategy defined once in prompt-cache.ts - Provider-specific config preserved: Bedrock gets 3 breakpoints + anchor, Anthropic family gets 2 breakpoints Preserved (untouched): - systemProviderOptions in all providers' streamText() calls - OpenAI Native promptCacheRetention (provider-level, not per-message) - Bedrock usePromptCache opt-in + supportsAwsPromptCache() 5,491 tests pass, 0 regressions.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary
RooMessagetype system that wraps AI SDKModelMessagewith metadata (timestamps, condense/truncation IDs), replacing the legacy Anthropic-formatApiMessageresult.response.messagesfrom the AI SDK for assistant message storage instead of manually reconstructing messages from stream events and injecting custom blocks (thinking,redacted_thinking,thoughtSignature)ModelMessageschemagetThoughtSignature,getRedactedThinkingBlocks,getReasoningDetails,getSummary) and their backing fields from all providersanthropicToRoo.ts) for loading old Anthropic-format conversation historiesTest plan
tsc --noEmit)🤖 Generated with Claude Code