Skip to content

CHANGELOG.md

Latest commit

 

History

History
102 lines (76 loc) · 4.99 KB

CHANGELOG.md

File metadata and controls

102 lines (76 loc) · 4.99 KB

Changelog

All notable changes to the session-protocol plugin documented here.

Format: Common Changelog + Keep a Changelog Versioning: Semantic Versioning

Unreleased

1.2.0.0 - 2026-02-06

Removed

  • Breaking: /save-session-protocol and /load-session-protocol slash commands - managing-session-continuity skill can be invoked directly if necessary
  • Command files: save-session-protocol.md and load-session-protocol.md

Changed

  • Improved assistant instruction in session start hook: If a session protocol exists provide complete workflow to use to the model when user wants to load a session protocol

1.1.0.0 - 2026-01-08

Changed

  • Breaking: Removed notification hook and related notification infrastructure from session-protocol plugin
  • Notification functionality moved to separate cross-platform-notification plugin for cleaner feature separation

Removed

  • Notification hook integration from hooks.json (moved to cross-platform-notification plugin)
  • PreCompact hook integration from hooks.json
  • hook-wrapper-with-notification.py: Unified notification wrapper script
  • precompact-session-protocol-reminder.py: PreCompact hook script for save reminders
  • claude_code_notifier.py: Cross-platform notification helper script

1.0.0.0 - 2025-11-27

First release of session continuity plugin.

Added

Core Functionality:

  • managing-session-continuity skill for session state management
  • /save-session-protocol command for capturing session context
  • /load-session-protocol command for restoring session context
  • WF1: Save Context - capture session state to sp.json with task consolidation
  • WF2: Load Context - restore session state from sp.json with validation

Task Management:

  • Task consolidation logic (3+ criteria decision matrix) for grouping related completed tasks
  • Consolidated task schema with consolidated bool and consolidated_count int fields
  • Session state check: skip Read if protocol loaded this session (overwrite), else merge
  • Task limit: last 5 individual completed + consolidated groups (no hard limit on groups)
  • Task categorization: BUGFIX, FEATURE, CONFIG, DOCS, TEST, REFACTOR
  • Priority levels: P1, P2, P3
  • File references in path:line format with relative or ~/ paths
  • Consolidation matrix: same feature/area, >5 tasks, no critical findings, >3 days age
  • Keep individual rules: critical bugs, unique pitfalls, recent (<3 days), referenced by pending

Git Integration:

  • Smart git integration: only check if .git exists, minimal commands for branch/commit
  • Git validation on load: simple commit comparison, branch drift warning
  • Graceful degradation when not a git repository

Data Format:

  • Privacy-safe JSON format: strip usernames from paths, no personal info
  • Minified JSON format requirement for token efficiency
  • ISO8601 UTC timestamps with Z suffix
  • Context blocks for architectural decisions, critical pitfalls, error patterns, plan refs

Hooks & Automation:

  • SessionStart hook with notification wrapper for automatic session detection
  • PreCompact hook with notification wrapper for session protocol reminders
  • Notification hook for cross-platform system notifications (Windows, macOS, Linux)
  • Hook-wrapper pattern for unified notification handling
  • Official Claude Code hook response schema with structured additionalContext

Token Optimization:

  • Token-optimized: 50% skill reduction vs traditional verbose markdown (~1200tk vs ~2400tk)
  • Workflow optimizations: task consolidation, conditional git, session state tracking
  • Optimization techniques: R1 (Remove Claude Knowledge), R2 (Compress Structure), R3 (Decision Matrices), R7 (Workflow Structure), R8A (Field Compression), R10 (Consistent Terminology)
  • SKILL.md: 271 lines vs ~500 traditional verbose (46% reduction)

Error Handling:

  • Error handling: graceful degradation for missing .git, invalid JSON errors
  • Clear error messages for debugging

Changed

  • Optimized SKILL.md to 271 lines vs ~500 traditional verbose (46% reduction)
  • Token efficiency: 50% reduction vs traditional markdown documentation (~1200tk vs ~2400tk)
  • Removed file validation (overhead without critical benefit)
  • Removed expensive git operations (commit count history)
  • Compressed structure: abbreviations (sp.json, pend, prog, w/, ts, mgmt), symbols (→, ≥, ≤)
  • Imperative format: "Read + parse" not "Use Read tool to read and then parse"
  • Decision matrix format for consolidation criteria
  • Workflow structure with numbered steps vs prose paragraphs