Skip to content

CHANGELOG.md

Latest commit

 

History

History
86 lines (62 loc) · 2.92 KB

CHANGELOG.md

File metadata and controls

86 lines (62 loc) · 2.92 KB

Changelog

All notable changes to Maestro will be documented in this file.

2.0.0 — 2026-04-26

⚡ Breaking

  • All packages bumped to v2.0.0 (extension, MCP server, @maestro/core)
  • All 25 skills bumped to v2.0.0

🆕 Added

Memory Layer

  • /capture command — save session summaries to .maestro/sessions/
  • /recap command — quick last-session recovery
  • /reflect command — effectiveness scorecard from audit data
  • decisions.jsonl — append-only decision log in .maestro/
  • audit.jsonl — command invocation audit trail with duration and cost
  • .maestro/ directory — persistent memory that survives sessions

Cost Estimation

  • Model pricing table for Claude, GPT-4, Gemini, o3, o4-mini (±20% accuracy)
  • Automatic cost tracking on every command invocation

MCP Tools

  • maestro_write_decision — append decisions via MCP
  • maestro_read_decisions — read decision history via MCP
  • maestro_read_audit — read audit trail via MCP

Extension

  • Automatic audit + decision emission after every command (wave and single-shot)
  • 3 new command palette entries (Capture, Reflect, Recap)
  • 3 new chat participant slash commands

🔧 Fixed

  • P0: npm publish blocker — MCP server now uses esbuild; @maestro/core bundled inline (no file: dependency at runtime)
  • P1: Dead import — removed invalid import { v4 as uuidv4 } from 'crypto' from wave-engine.ts
  • P2: 5.45MB bundle — replaced js-tiktoken (4MB vocab) with character-count heuristic; extension reduced to 124KB (98% smaller)

🏗️ Changed

  • Context resolution now checks .maestro/context.md first, falls back to .maestro.md (backward compatible)
  • agent-workflow skill updated with .maestro/ directory awareness and decision history protocol
  • All READMEs updated (main, extension, MCP server)
  • Multi-provider build: 10 providers × 25 skills = 250 skill copies

🧪 Testing

  • Added vitest to @maestro/core
  • 37 unit tests: context-utils (11), token-estimator (6), cost-estimator (9), decisions (8), audit (3)
  • All tests pass

1.4.2 — 2026-04-12

Added

  • Wave execution engine for multi-phase commands
  • MCP wave tools (start, advance, status)
  • Context slicer for token optimization
  • Workspace indexer for file dependency tracking

Changed

  • Extension sidebar redesigned with token budget display
  • MCP server supports HTTP transport

1.3.1 — 2026-04-04

Added

  • 21 commands across 4 categories
  • agent-workflow core skill with 7 domain references
  • Multi-provider skill sync (10 providers)
  • Zero-defect mode with auto-inject
  • @maestro chat participant
  • .maestro.md context protocol
  • MCP server with stdio and HTTP transport