Skip to content

CHANGELOG.md

Latest commit

 

History

History
117 lines (90 loc) · 4.79 KB

CHANGELOG.md

File metadata and controls

117 lines (90 loc) · 4.79 KB

Changelog

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

The format is based on Keep a Changelog.

[3.0.0] — 2026-04-13

Major UX refactor. Tool surface collapsed to 3, sessions added, visualization made on-demand, approval moved off the filesystem. See MIGRATION_v2_to_v3.md for the full migration guide.

Breaking Changes

  • Tool renames:
    • AoT-light → AoT-fast
    • AoT → AoT-full
  • Tools removed (folded into other tools):
    • generate_visualization → set viz: true on AoT-fast / AoT-full
    • check_approval → atomcommands subcommand "check_approval"
    • export_graph → atomcommands subcommand "export"
  • Server flags removed:
    • --no-viz, --no-approval, --viz (boolean) → use --viz auto|always|never
  • Config fields removed: vizEnabled, approvalEnabled → replaced by vizMode: 'auto' | 'always' | 'never'
  • processAtom response shape: now includes sessionId; empty collection fields and terminationStatus are omitted when not meaningful (callers that snapshot the JSON should expect a leaner shape)

Added

  • Sessions: atom state scoped per-session. Default session "default". Two reasoning problems in one process no longer collide.
  • atomcommands subcommands: new_session, switch_session, list_sessions, reset_session
  • Auto-archive: session marked completed when reasoning terminates
  • Auto-spawn: next zero-dep atom in a completed session spawns a fresh default-N session automatically
  • HTTP approval callback: local 127.0.0.1 listener on an ephemeral port. Browser POSTs approval JSON back via XHR. Falls back to file polling on the configured downloads dir if the POST fails.
  • viz: true param on AoT calls renders the D3 graph and opens the browser. Server flag --viz auto|always|never controls overall policy.
  • sessionId param on AoT calls targets a specific session (auto-creates if unknown).

Changed

  • Tool count: 6 → 3 (AoT-fast, AoT-full, atomcommands)
  • Default response payload: ~12 fields → ~6 (empty arrays, null fields, and unactionable termination status omitted)
  • Tool descriptions tightened with explicit fast/full decision rule and planning-mode visualization heuristic
  • approval.ts uses os.homedir() for cross-platform Downloads dir detection (was process.env.HOME)

Migration

  • Replace mcp__atom-of-thoughts__AoT-light with mcp__atom-of-thoughts__AoT-fast
  • Replace mcp__atom-of-thoughts__AoT with mcp__atom-of-thoughts__AoT-full
  • Replace mcp__atom-of-thoughts__generate_visualization(...) with mcp__atom-of-thoughts__AoT-full({..., viz: true})
  • Replace mcp__atom-of-thoughts__check_approval() with mcp__atom-of-thoughts__atomcommands({command: "check_approval"})
  • Replace mcp__atom-of-thoughts__export_graph() with mcp__atom-of-thoughts__atomcommands({command: "export"})
  • Replace --no-viz with --viz never, --no-approval is gone (approval is always on; control viz via --viz)

Tests

  • 121 → 165 tests passing across 12 test files
  • New: tests/sessions.test.ts, tests/approval-server.test.ts, tests/payload-shape.test.ts

[2.1.0] — 2026-03-11

Changed

  • Visualization off by default — pass --viz to enable (was --no-viz to disable)
  • Lenient validation — partial atom inputs accepted; missing fields get sensible defaults
  • Shorter tool descriptions — improved display in Claude Code tool listings
  • Removed isError from non-error response shapes for cleaner MCP compliance

Fixed

  • Tests updated to match new defaults and validation behavior

[2.0.0] — 2026-02-13

Initial public release.

Added

  • AoT — Full decomposition with up to 5 depth levels and confidence tracking
  • AoT-light — Quick analysis mode (max 3 levels, no visualization)
  • atomcommands — Advanced atom manipulation (contract, merge, prune, split, reweight)
  • export_graph — Export reasoning graphs as JSON or Mermaid
  • generate_visualization — Interactive D3.js force-directed graph with approve/reject workflow
  • check_approval — Poll user approval status from the visualization UI
  • 5 atom types with confidence thresholds: Premise (0.9), Reasoning (0.8), Hypothesis (0.7), Verification (0.85), Conclusion (0.9)
  • CLI flags: --mode, --max-depth, --no-viz, --no-approval, --output-dir, --downloads-dir
  • Docker support with multi-stage build
  • Smithery deployment configuration
  • GitHub Actions CI across Node 18, 20, 22
  • 121 tests covering all tools, atom types, and edge cases

Technical

  • Built on @modelcontextprotocol/sdk ^1.24.0 (MCP protocol 2025-03-26)
  • TypeScript strict mode, ES2022 target
  • Vitest test runner
  • Bundled D3.js v7 for offline visualization