fix(conversation-memory): move vitest.config.ts to plugin directory to reduce resource usage

[baleen37]

Summary

• Move vitest.config.ts from worktree root to plugins/conversation-memory/ directory
• Update exclude pattern to use recursive glob: **/node_modules/**

Problem

Vitest was consuming excessive resources when running tests:

• Scanning and including 144 test files from node_modules/zod
• Collection time: 10.60s
• Max memory: 203MB

Root Cause

• vitest.config.ts was in worktree root with include: ['**/*.test.ts']
• exclude: ['node_modules'] only excluded root node_modules, not plugin's node_modules

Solution

• Move config to plugin directory (plugins/conversation-memory/)
• Update exclude pattern to ['**/node_modules/**']

Performance Improvement

Metric	Before	After	Improvement
Collection time	10.60s	1.90s	5.6x faster
Total duration	1.53s	0.64s	2.4x faster
Max memory	203MB	170MB	16% reduction

Test plan

• All tests pass (585 tests)
• No zod tests are included in test run
• Resource usage significantly reduced

🤖 Generated with Claude Code

Summary by CodeRabbit

Release Notes

• New Features

  • Redesigned conversation memory system (v3) with batch-based observation extraction via LLM.
  • Added three-tier progressive disclosure: compact search results, full observation details, and raw conversation transcripts.
  • Implemented new hook-based architecture replacing background observer with session-start injection, tool-use tracking, and session-end extraction.

• Database Changes

  • Migrated to simplified schema featuring pending_events and observations tables with vector and full-text search indices.

• Refactoring

  • Removed polling daemon and legacy CLI commands. Observation processing now event-driven through hooks.

[coderabbitai]

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

Introduces a comprehensive v3 redesign of conversation-memory from an observer-polling daemon to a hook-based async batch processing architecture. Replaces legacy components with new modules: v3 database schema (pending_events, observations, vector embeddings, FTS), compression utilities, LLM batch extraction, three hooks (PostToolUse, Stop, SessionStart), and a simplified MCP server with three tools (search, get_observations, read). Removes extensive legacy code (observer, indexer, sync, search, parser, and associated CLI commands and tests).

Changes

Cohort / File(s)	Summary
Design & Configuration docs/plans/2026-02-09-conversation-memory-v3-design.md, plugins/conversation-memory/vitest.config.ts	Comprehensive v3 architecture design document and updated test config for nested node_modules exclusion.
Hooks Configuration plugins/conversation-memory/hooks/hooks.json, plugins/conversation-memory/hooks/hooks.test.ts	Removed legacy sync/observer startup hooks; added comprehensive schema validation tests for remaining SessionStart/PostToolUse/Stop hooks with v3 implementation references.
CLI - Legacy Removal plugins/conversation-memory/src/cli/index-cli.ts, plugins/conversation-memory/src/cli/inject-cli.ts, plugins/conversation-memory/src/cli/observe-cli.ts, plugins/conversation-memory/src/cli/observer-cli.ts, plugins/conversation-memory/src/cli/search-cli.ts, plugins/conversation-memory/src/cli/sync-cli.ts	Removed extensive CLI commands (inject, observe, observer, search, sync) and multi-command interface; simplified to MCP/hook-based operations only.
CLI Tests - Legacy Removal plugins/conversation-memory/src/cli/search-cli.test.ts, plugins/conversation-memory/src/cli/show-cli.test.ts, plugins/conversation-memory/src/cli/stats-cli.test.ts, plugins/conversation-memory/src/cli/sync-cli.test.ts	Removed test suites for legacy CLI components (search, show, stats, sync).
Database Layer V3 plugins/conversation-memory/src/core/db.v3.ts, plugins/conversation-memory/src/core/db.v3.test.ts	New clean-slate v3 database with tables: pending_events, observations, vec_observations (vector embeddings), observations_fts (full-text search); includes CRUD operations, vector/FTS integration, and comprehensive test coverage.
Database Layer Legacy plugins/conversation-memory/src/core/db.ts, plugins/conversation-memory/src/core/db.test.ts	Removed entire v2 database implementation with exchanges, tool_calls, vector tables, migrations, and all associated CRUD operations and tests.
Tool Compression V3 plugins/conversation-memory/src/core/compress.ts, plugins/conversation-memory/src/core/compress.test.ts	New module for rule-based compression of tool outputs (Read, Edit, Write, Bash, Grep, WebSearch, WebFetch) into concise strings for pending_events storage.
LLM Batch Extraction plugins/conversation-memory/src/core/llm/batch-extract-prompt.ts, plugins/conversation-memory/src/core/llm/batch-extract-prompt.test.ts	New module for batch-processing compressed tool events via LLM to extract observations with deduplication using previous observations as context.
LLM Configuration & Provider plugins/conversation-memory/src/core/llm/config.ts, plugins/conversation-memory/src/core/llm/config.test.ts, plugins/conversation-memory/src/core/llm/index.ts, plugins/conversation-memory/src/core/llm/index.test.ts	Simplified LLM config from multi-key Gemini to single apiKey model; removed RoundRobinProvider; added batch-extract-prompt exports.
LLM Provider Legacy plugins/conversation-memory/src/core/llm/round-robin-provider.ts, plugins/conversation-memory/src/core/llm/round-robin-provider.test.ts	Removed RoundRobinProvider for distributing requests across multiple API keys.
Observations V3 plugins/conversation-memory/src/core/observations.v3.ts, plugins/conversation-memory/src/core/observations.v3.test.ts	New module providing CRUD and vector-search capabilities for observations with embedding generation, project/session filtering, and vector similarity search.
Observations Legacy plugins/conversation-memory/src/core/observations.ts, plugins/conversation-memory/src/core/observations.test.ts	Removed v2 observation management with query helpers, validation, and database integration.
Search V3 plugins/conversation-memory/src/core/search.v3.ts, plugins/conversation-memory/src/core/search.v3.test.ts	New observation-only search supporting vector similarity, full-text search, and hybrid modes with ISO date validation, project/file filtering, and result deduplication.
Search Legacy plugins/conversation-memory/src/core/search.ts, plugins/conversation-memory/src/core/search.test.ts	Removed v2 conversation/observation search, recency boosting, multi-concept search, and associated formatting logic.
Read Tool plugins/conversation-memory/src/core/read.ts, plugins/conversation-memory/src/core/read.test.ts	New module for reading raw JSONL conversation files with Markdown formatting, optional pagination (startLine/endLine), and metadata/sidechain support.
Parser & Sync Legacy plugins/conversation-memory/src/core/parser.ts, plugins/conversation-memory/src/core/parser-*.test.ts, plugins/conversation-memory/src/core/sync.ts, plugins/conversation-memory/src/core/sync.test.ts, plugins/conversation-memory/src/core/indexer.ts, plugins/conversation-memory/src/core/indexer.test.ts	Removed v2 conversation parser with exclusion logic, exchange construction, tool-call handling; removed sync/archive pipeline; removed indexer with batch processing and embedding generation.
Inject & Session Summary Legacy plugins/conversation-memory/src/core/inject.ts, plugins/conversation-memory/src/core/session-summary.ts, plugins/conversation-memory/src/core/session-summary.test.ts, plugins/conversation-memory/src/core/tool-compress.ts, plugins/conversation-memory/src/core/tool-compress.test.ts	Removed v2 context injection, session summary management, and tool-call formatting with per-tool formatters.
Observer & Verification Legacy plugins/conversation-memory/src/core/observer.ts, plugins/conversation-memory/src/core/observation-prompt.ts, plugins/conversation-memory/src/core/observation-prompt.test.ts, plugins/conversation-memory/src/core/verify.ts	Removed v2 in-process observer daemon with polling loop, session lifecycle, and prompt building; removed index verification and repair utilities.
Hooks - PostToolUse plugins/conversation-memory/src/hooks/post-tool-use.ts, plugins/conversation-memory/src/hooks/post-tool-use.test.ts	New hook that compresses tool data and inserts pending events into database for later batch processing.
Hooks - Stop plugins/conversation-memory/src/hooks/stop.ts, plugins/conversation-memory/src/hooks/stop.test.ts	New hook that batches pending events (10-20), calls LLM for observation extraction with previous batch context, and persists observations with embeddings.
Hooks - SessionStart plugins/conversation-memory/src/hooks/session-start.ts, plugins/conversation-memory/src/hooks/session-start.test.ts	New hook that queries recent observations, applies token budget, formats as Markdown, and injects as session context with recency/project filtering.
MCP Server plugins/conversation-memory/src/mcp/server.ts, plugins/conversation-memory/src/mcp/server.test.ts, plugins/conversation-memory/src/mcp/server.v3.test.ts	Replaced v2 multi-tool MCP server with v3 three-tool architecture: search (observations), get_observations (by IDs), read (raw JSONL); updated schemas, removed multi-concept path, bumped version to 3.0.0; added comprehensive v3 test suite.
Types plugins/conversation-memory/src/core/types.ts	Removed similarity property from CompactSearchResult.

Sequence Diagram(s)

sequenceDiagram
    participant User as User/Claude
    participant PostTool as PostToolUse Hook
    participant DB as Database<br/>(pending_events)
    participant Stop as Stop Hook
    participant LLM as LLM Provider
    participant Obs as Database<br/>(observations)
    participant Start as SessionStart Hook
    participant Context as Context Injection

    User->>PostTool: Tool invocation completes
    PostTool->>PostTool: compressToolData()
    PostTool->>DB: insertPendingEventV3
    DB-->>PostTool: event stored

    rect rgba(100, 150, 200, 0.5)
    Note over Stop,Obs: End of session
    Stop->>DB: getPendingEventsV3(sessionId)
    DB-->>Stop: pending events list
    Stop->>Stop: createBatches (10-20 events)
    
    loop For each batch
        Stop->>Stop: Build batch prompt w/ prev observations
        Stop->>LLM: extractObservationsFromBatch()
        LLM-->>Stop: extracted observations
        Stop->>Obs: insertObservationV3 (with embedding)
        Obs-->>Stop: observation stored
        Stop->>Stop: Accumulate for next batch context
    end
    end

    rect rgba(100, 200, 150, 0.5)
    Note over Start,Context: New session starts
    Start->>Obs: searchObservationsV3(recent)
    Obs-->>Start: observations (last 7 days)
    Start->>Start: Apply token budget
    Start->>Start: formatObservation() to Markdown
    Start->>Context: Inject markdown context
    Context-->>User: Context available in session
    end

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• PR #452: Implements the same v3 observation-based redesign with identical hook-based architecture and module restructuring across database, observations, search, and MCP server layers.
• PR #449: Adds tool-call compression (formatToolSummary + compressed_tool_summary) in v2 indexing stack; this PR replaces that approach with new compressToolData module and hook-based pending event processing.
• PR #425: Restores conversation-memory plugin; directly conflicts with this PR's v3 redesign as both replace the same plugin files and implementations.

Poem

🐰 From polling daemons to hooks we hop,\br
Pending events batch, observations pop!$br
Compress the tools, let LLMs extract,$br
Token-budgeted context, perfectly packed.$br
Observer's gone—hooks now do the dance! 🎉

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch chore/claude-mem3

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}