SDK runner: close message processing parity gaps with Pi embedded agent#12
Closed
SDK runner: close message processing parity gaps with Pi embedded agent#12
Conversation
Implement a comprehensive scoring system that evaluates whether experiences should become long-term memories based on five weighted factors: - **Novelty**: Is this new information or repetition? - **Impact**: Does this change understanding, behavior, or system state? - **Relational**: Does this connect to known entities/people/projects? - **Temporal**: Is this time-sensitive or evergreen? - **User Intent**: Was this explicitly marked as important? Features: - Configurable weights for each factor (normalized to sum to 1.0) - Named threshold profiles: balanced, aggressive, conservative, minimal - Override rules for specific tools (glob-pattern matching) - evaluateMemoryRelevance() as primary entry point - shouldCapture(), shouldPersistToGraph(), shouldUseLlmEval() helpers - Backward-compatible evaluateHeuristic() now delegates to scoring system - New evaluateHeuristicDetailed() returns full scoring breakdown - 37 comprehensive tests covering all factors and edge cases Files: - extensions/meridia/src/meridia/scoring/types.ts - extensions/meridia/src/meridia/scoring/config.ts - extensions/meridia/src/meridia/scoring/factors.ts - extensions/meridia/src/meridia/scoring/index.ts - extensions/meridia/src/meridia/scoring/scoring.test.ts - extensions/meridia/src/meridia/evaluate.ts (updated)
feat(meridia): multi-factor memory relevance scoring system
feat: add configurable subsystem debug log suppression
- Gap 1 (HIGH): Add streaming block replies during the event loop via blockReplyBreak/blockReplyChunking params. Supports "message_end" (flush at message boundaries) and "text_end" (per-chunk) modes, matching Pi embedded's streaming delivery. Final onBlockReply is skipped when streaming was active to avoid duplicate delivery. - Gap 2 (MEDIUM): Add onReasoningStream callback support. Extract thinking/reasoning text from SDK events classified as "system" and emit via the new callback, enabling reasoning display in UIs. - Gap 3 (MEDIUM): Reset chunks/assistantSoFar at message boundaries so onPartialReply reflects only the current message, not accumulated text from all prior turns. - Gap 5 (LOW-MED): Add enforceFinalTag support. When enabled, only content inside <final>...</final> tags is returned, preventing intermediate narration leakage from non-Claude models. Adds shared extractFinalTagContent() utility. - Gap 6 (LOW-MED): Map all payload fields (mediaUrl, mediaUrls, replyToId, replyToTag, replyToCurrent, audioAsVoice) in adaptSdkResultToPiResult to prevent silent field loss. Wire all new params through the full chain: executor.ts → AgentRuntimeRunParams → sdk-agent-runtime → sdk-runner-adapter → sdk-runner https://claude.ai/code/session_013FRzxnpzCwncubghL7GGCF
…-processing-djLHt
Resolve merge conflicts in meridia scoring module by taking origin/main's refactored multi-factor scoring system (scorer.ts, defaults.ts, updated types/factors/index). Remove HEAD-only scoring/config.ts and scoring.test.ts that used the old API (origin/main has proper tests in factors.test.ts and scorer.test.ts). Co-Authored-By: Claude Opus 4.6 <[email protected]>
dgarson
added a commit
that referenced
this pull request
Feb 7, 2026
…#208) * feat: added work-queue workflow workers and a worker manager, integrate w/cron * fix: address 12 logic flaws in workflow engine, adapter, phases, and cron types - engine: capture phase before clobbering in catch handler (#1) - engine: remove dead `state.plan ?? plan` fallback (#2) - engine: mark workflow failed when all execution nodes fail (#6) - adapter: fix retry count off-by-one (attemptNumber vs retryCount) (#3) - adapter: clean up abort listener in sleep on timeout (#10) - discover: fix batch/entries index mismatch when spawns fail (#4) - execute: add cycle detection to topological sort (#5) - review: add autoApproved flag to distinguish fallback approvals (#7) - plan: add comment clarifying sessionKey reuse across repair attempts (#8) - decompose: remove unused model/maxPhases/maxTasksPerPhase/maxSubtasksPerTask opts (#9) - types: add autoApproved to ReviewIteration (#7) - cron/state: use discriminated union for CronEvent (CronJobEvent | CronHealthEvent) (#12) - tests: add WorkflowWorkerAdapter test suite (8 tests) (#11) https://claude.ai/code/session_01L8kquwpmUh5zmU9S4MHgPu --------- Co-authored-by: Claude <[email protected]>
dgarson
added a commit
that referenced
this pull request
Feb 7, 2026
…cron types (openclaw#217) - engine: capture phase before clobbering in catch handler (#1) - engine: remove dead `state.plan ?? plan` fallback (#2) - engine: mark workflow failed when all execution nodes fail (#6) - adapter: fix retry count off-by-one (attemptNumber vs retryCount) (#3) - adapter: clean up abort listener in sleep on timeout (#10) - discover: fix batch/entries index mismatch when spawns fail (#4) - execute: add cycle detection to topological sort (#5) - review: add autoApproved flag to distinguish fallback approvals (#7) - plan: add comment clarifying sessionKey reuse across repair attempts (#8) - decompose: remove unused model/maxPhases/maxTasksPerPhase/maxSubtasksPerTask opts (#9) - types: add autoApproved to ReviewIteration (#7) - cron/state: use discriminated union for CronEvent (CronJobEvent | CronHealthEvent) (#12) - tests: add WorkflowWorkerAdapter test suite (8 tests) (#11) https://claude.ai/code/session_01L8kquwpmUh5zmU9S4MHgPu Co-authored-by: Claude <[email protected]>
Owner
Author
|
Already merged to main via git merge. Closing stale PR. |
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
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.
Gap 1 (HIGH): Add streaming block replies during the event loop via blockReplyBreak/blockReplyChunking params. Supports "message_end" (flush at message boundaries) and "text_end" (per-chunk) modes, matching Pi embedded's streaming delivery. Final onBlockReply is skipped when streaming was active to avoid duplicate delivery.
Gap 2 (MEDIUM): Add onReasoningStream callback support. Extract thinking/reasoning text from SDK events classified as "system" and emit via the new callback, enabling reasoning display in UIs.
Gap 3 (MEDIUM): Reset chunks/assistantSoFar at message boundaries so onPartialReply reflects only the current message, not accumulated text from all prior turns.
Gap 5 (LOW-MED): Add enforceFinalTag support. When enabled, only content inside ... tags is returned, preventing intermediate narration leakage from non-Claude models. Adds shared extractFinalTagContent() utility.
Gap 6 (LOW-MED): Map all payload fields (mediaUrl, mediaUrls, replyToId, replyToTag, replyToCurrent, audioAsVoice) in adaptSdkResultToPiResult to prevent silent field loss.
Wire all new params through the full chain:
executor.ts → AgentRuntimeRunParams → sdk-agent-runtime → sdk-runner-adapter → sdk-runner
https://claude.ai/code/session_013FRzxnpzCwncubghL7GGCF