Release v4.3.0: Progressive Disclosure Context
Major Feature: - Progressive Disclosure Context: 3-layer memory retrieval system - Layer 1 (Index): Observation titles, token costs, type indicators - Layer 2 (Details): Full narratives via MCP search on-demand - Layer 3 (Perfect Recall): Source code and original transcripts - Context hook displays observations in table format - Type indicators: 🔴 critical, 🟤 decision, 🔵 informational - Token counts help Claude decide: fetch details vs read code Added: - Agent Skills documentation and version bump management skill - Memory toggle feature planning document Changed: - Enhanced session summary handling and timeline rendering - Context token cost increased from ~800 to ~2,500 tokens Fixed: - Removed hardcoded macOS-specific paths (fixes #23) - Cross-platform path detection improvements Files Updated: - package.json, marketplace.json, plugin.json: version 4.3.0 - CLAUDE.md: version history and current version updated - README.md: removed experimental section, integrated feature - CHANGELOG.md: comprehensive v4.3.0 release entry - Built all hooks and worker service 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
@@ -8,6 +8,33 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
|
||||
## [Unreleased]
|
||||
|
||||
|
||||
## [4.3.0] - 2025-10-25
|
||||
|
||||
### Added
|
||||
- **Progressive Disclosure Context**: Enhanced context hook with layered memory retrieval system
|
||||
- Layer 1 (Index): Observation titles, token costs, and type indicators at session start
|
||||
- Layer 2 (Details): Full narratives retrieved on-demand via MCP search
|
||||
- Layer 3 (Perfect Recall): Source code and original transcripts
|
||||
- Context hook now displays observations in table format with ID, timestamp, type indicator, title, and token count
|
||||
- Type indicators: 🔴 (critical/gotcha), 🟤 (decision), 🔵 (informational/how-it-works)
|
||||
- Progressive disclosure instructions guide Claude on when to fetch full observation details vs. reading code
|
||||
- Token counts (~200-500 per observation) help Claude make informed retrieval decisions
|
||||
- **Agent Skills documentation**: Added comprehensive documentation on creating and using Claude Code agent skills
|
||||
- **Version bump skill**: Added automated version bump management skill for streamlined releases
|
||||
- **Memory toggle feature planning**: Added design document for future pause/resume recording capability
|
||||
|
||||
### Changed
|
||||
- **Enhanced session summary handling**: Improved timeline rendering and summary organization
|
||||
- **Improved context hook output**: Added structured timeline with session grouping and observation details
|
||||
- **Context token cost**: Increased from ~800 tokens to ~2,500 tokens for richer observation index
|
||||
|
||||
### Fixed
|
||||
- **Cross-platform path detection**: Removed hardcoded macOS-specific paths for project and Claude Code executable (fixes #23)
|
||||
- Removed hardcoded paths in context hook, worker service, and SDK integration
|
||||
- Now uses dynamic path resolution for cross-platform compatibility
|
||||
- Affects: `src/hooks/context.ts`, `src/services/worker-service.ts`, `src/sdk/worker.ts`
|
||||
|
||||
|
||||
## [4.2.11] - 2025-10-25
|
||||
|
||||
### Fixed
|
||||
|
||||
Reference in New Issue
Block a user