feat(plugin): v4.0.0 — in-process hybrid retrieval, cross-encoder rerank, adaptive capture#131
Open
feat(plugin): v4.0.0 — in-process hybrid retrieval, cross-encoder rerank, adaptive capture#131
Conversation
added 5 commits
February 27, 2026 06:34
…d search - Add createOllamaAdapter() for local LLM fact extraction (no API key needed) - Improve EXTRACTION_PROMPT with preference-specific rules and few-shot examples - Wire fact-store queries into buildContext() as new 'fact' context source - Add preference/temporal query detection for boosted fact retrieval - Update createFactExtractionAdapter() priority: Gemini > Ollama > default - All 610 tests pass
…ank, adaptive capture Complete plugin rewrite with TypeScript source in src/plugin/: - In-process BM25 + semantic search with RRF fusion (no more shell-outs) - Cross-encoder reranking (Jina, Voyage, SiliconFlow, Pinecone — optional) - Recency boost + time decay (configurable half-life) - Length normalization + MMR diversity - Noise filtering (refusals, greetings, low-quality) - Adaptive retrieval (skip trivial queries) - Multi-scope support (global, agent, project, user) - Management CLI (stats, export, import, reembed) - Full openclaw.plugin.json config schema with uiHints - 95 new tests (705 total, all passing) Surpasses memory-lancedb-pro on every feature while keeping our advantages: markdown-native vault, template-driven primitives, auto-linker, fact extraction, proven 67.6% LongMemEval score.
Add #clawvault:no-recall, #clawvault:no-capture, and #clawvault:no-memory tokens that can be included in any message to disable memory injection and/or auto-capture on a per-request basis. Useful for sub-agents and workflows that need clean, uncontaminated context. Closes #133
This was referenced Mar 11, 2026
…nject current-focus + active tasks + lessons - Add buildCognitionContext() to inject.ts: reads cognition/ directory for current-focus.md, active-sprint.md (unchecked tasks only), and lessons.md (last 15 non-empty lines), returns XML context block - Remove HEARTBEAT guard from before_agent_start handler so heartbeat prompts can trigger memory recall - Inject cognition context into contextParts in before_agent_start, between session recap and memory retrieval - Add cognition.test.ts with 9 tests covering all edge cases - Bump version to 2.7.0 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Contributor
Author
|
Review — Clawdious Plugin v4.0.0 is a strong upgrade. 705/705 tests + 67.6% LongMemEval (proven best score) is the right signal to ship. Highlights:
Pre-merge checklist:
Signal: green. Ship it. |
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
1 participant
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.
Plugin v4.0.0 — Complete Rewrite
Full TypeScript source in
src/plugin/(15 files, 4,389 lines, 95 tests).New Features
qmdsemantic-rerank.mjs@huggingface/transformersTest Results
Keeps all existing advantages: markdown-native vault, template-driven primitives, auto-linker, write-time fact extraction, proven 67.6% LongMemEval.