v1.6.3

Changelog

All notable changes to this project will be documented in this file. See standard-version for commit guidelines.

[1.6.3] - 2026-05-23

Changed

• Refactored cursor/plugins merged skills into single cohesive bodies (no append-only Imported from sections or -v2 parallel skills).
• Integrated thermo-nuclear review as strict maintainability mode in code-review-and-quality and code-reviewer.
• Renamed continuous-learning-v2 to canonical continuous-learning with unified instinct + workspace memory workflow.
• Import script now treats merges as native-only (re-import never appends duplicate skill/agent sections).
• Coverage verifier recognizes native integration markers for merged agents.

Added

• Full cursor-team-kit agent import set (ci-watcher, compatibility agents, agents-memory-updater, poteto-agent, etc.) wired in PLUGIN_MAP.

[0.6.0] - 2026-03-30

Added

• Added four new specialists to close high-priority coverage gaps:
  • mobile-developer
  • data-engineer
  • aws-architect
  • agent-developer
• Added coordinated registry and reference updates so the published agent inventory now reflects the actual 32-agent system.

Changed

• Normalized canonical and generated agent inventories across content/, .claude-plugin/, .opencode/, plugins/ai-eng-system/, packages/core/, and packages/toolkit/.
• Updated marketplace metadata, installation docs, and agent registries to use the current 32-agent count.
• Bumped the coordinated package version to 0.6.0 for @ai-eng-system/core, @ai-eng-system/toolkit, and @ai-eng-system/cli.

Fixed

• Fixed stale subagent orchestration references so routing guidance now points to the correct current specialists, including api-builder-enhanced.

[0.5.10] - 2026-03-27

Added

• Published the first coordinated three-package release:
  • @ai-eng-system/core
  • @ai-eng-system/toolkit
  • @ai-eng-system/cli
• Added @ai-eng-system/toolkit as a real workspace package built from generated .claude-plugin, .opencode, and marketplace plugin artifacts.
• Added stable toolkit path helpers for consumers that need packaged asset locations.
• Added the new /ai-eng/simplify command backed by the namespaced skill at skills/ai-eng/simplify/SKILL.md.

Changed

• Standardized release automation around the coordinated OIDC workflow in .github/workflows/publish-all-oidc.yml.
• Preserved namespaced skill paths such as ai-eng/simplify and workflow/ralph-wiggum across generated Claude, OpenCode, marketplace, and toolkit outputs.
• Updated installation, marketplace, publishing, and package-role documentation to reflect the three-package model.

Fixed

• Fixed trusted-publishing workflow issues involving workspace lockfiles, toolkit artifact generation in CI, and CLI publish manifest recursion.
• Fixed release workflow validation to use a release-safe CLI smoke-test suite.
• Fixed generated skill packaging so nested skill directories are no longer flattened incorrectly.

0.4.7 (2026-01-19)

0.4.4 (2026-01-17)

⚠ BREAKING CHANGES

• Model must now be explicitly configured in config.yaml
  or opencode.model. No automatic fallback to claude-3-5-sonnet-latest.
• None
  This addresses all requirements from issue analysis and provides
  a complete clean reinstall solution with working hooks.

Features

• ai-eng ralph CLI with timeout and model config (5562fc3)
• apply prompt optimization improvements across all agents and commands (550139e)
• v0.3.0 - Complete clean reinstall and hooks implementation (42ec81c)
• v0.4.0 CLI improvements and bug fixes (2561006)

Bug Fixes

• add @opentui/core dependency for TUI (2ce7eee)
• address pre-existing issues (5e470ff)
• Correct OpenCode messageID format and remove invalid config keys (08c2fe6)
• Resolve critical CLI reliability issues (678ebe5)
• resolve OpenCode session creation with proper server lifecycle management (e90f64f)
• use process.cwd() for working directory (6944987)

Build System

• add CLI to dist + remove model fallback (830eed5)

0.4.3 (2026-01-17)

⚠ BREAKING CHANGES

• Model must now be explicitly configured in config.yaml
  or opencode.model. No automatic fallback to claude-3-5-sonnet-latest.
• None
  This addresses all requirements from issue analysis and provides
  a complete clean reinstall solution with working hooks.

Features

• ai-eng ralph CLI with timeout and model config (5562fc3)
• apply prompt optimization improvements across all agents and commands (550139e)
• v0.3.0 - Complete clean reinstall and hooks implementation (42ec81c)
• v0.4.0 CLI improvements and bug fixes (2561006)

Bug Fixes

• add @opentui/core dependency for TUI (2ce7eee)
• address pre-existing issues (5e470ff)
• Correct OpenCode messageID format and remove invalid config keys (08c2fe6)
• Resolve critical CLI reliability issues (678ebe5)
• resolve OpenCode session creation with proper server lifecycle management (e90f64f)
• use process.cwd() for working directory (6944987)

Build System

• add CLI to dist + remove model fallback (830eed5)

[0.2.4] - 2026-01-12

Fixed

• OpenCode SDK Compatibility: Upgraded @opencode-ai/plugin from 1.0.218 to 1.1.13 to resolve messageID validation errors with OpenCode 1.1.13 server
• Session Management: SDK v1.1.13 fixes API response format validation issues that prevented proper session creation and closure

Changed

• Dependencies: Updated OpenCode SDK to 1.1.13 for compatibility with latest OpenCode CLI server
• Build: Regenerated bun.lock with updated dependency tree

Technical Details

The OpenCode 1.1.x server introduced API response format changes that were incompatible with SDK 1.0.x clients. The error manifested as:

Invalid string: must start with "msg"

Upgrading to SDK 1.1.13 resolves this without any breaking API changes to ai-eng-ralph-cli.

Verification

• ✅ Clean build successful with all 662 packages
• ✅ Tested with fleettools project - session created and closed successfully
• ✅ No source code changes required - SDK API remained compatible

0.2.3 (2026-01-10)

⚠ BREAKING CHANGES

• Default behavior is now autonomous. Use --refine-each-phase for original interactive behavior.

Bug Fixes

• ralph-wiggum autonomous execution between phases (16f9c1c)
• ralph-wiggum autonomous looping issue with continuous phase execution (53241c7)
• resolve plugin installation issues and sync commands (1bfc229)

[0.2.2] - 2026-01-07

Fixed

• Ralph Wiggum Autonomous Looping: Command no longer stops between workflow phases, ensuring continuous iteration
• Prompt Refinement Optimization: Only performs prompt-refinement once at Phase 0, storing refined context for subsequent phases
• Checkpoint Compatibility: Maintains full backward compatibility with existing checkpoint system
• Performance Enhancement: Reduced redundant prompt processing, improving workflow execution speed

Added

• --refine-each-phase Flag: New optional flag for interactive mode that refines prompts at each phase (default: disabled)
• Explicit Continuation Instructions: Added clear phase transition guidance to prevent workflow interruption
• Context Persistence: Refined context is now stored and reused across all phases for consistency

Changed

• Phase Transition Logic: Improved flow between Research → Specify → Plan → Work → Review phases
• Documentation: Updated ralph-wiggum.md with comprehensive usage examples and troubleshooting guide
• Command Structure: Enhanced from 761 to 789 lines with additional error handling and edge case coverage

Technical Details

The issue was caused by prompt-refinement being invoked at every phase transition, creating natural stopping points in the autonomous workflow. The fix involves:

1. Single Refinement: Prompt-refinement occurs only once at initialization (Phase 0)
2. Context Storage: Refined prompts are cached and reused for all subsequent phases
3. Continuation Signals: Explicit "continue to next phase" instructions prevent workflow interruption
4. Fallback Mode: --refine-each-phase flag allows original behavior for interactive use cases

Backward Compatibility

• All existing ralph-wiggum usage patterns remain unchanged
• Checkpoint system fully compatible with new workflow
• Default behavior provides the fix, no configuration required
• Optional flag available for users who prefer per-phase refinement

0.1.1 (2026-01-06)

Features

• integrate Ralph Wiggum skill and remove CLI executor (61c56be)

[0.0.14] - 2026-01-01

Fixed

• Documentation Accuracy: Corrected agent count from 29 to 28 across all documentation
• Skill Count: Corrected skill file count from 13 to 7 (SKILL.md files only)
• TODO.md Cleanup: Reorganized completed tasks, updated version references

Changed

• Version Consistency: All documentation now reflects accurate counts
• TODO.md: Moved completed High Priority items to Completed Tasks section

Infrastructure

• Build: Stable
• Commands: 17 total
• Agents: 28 total
• Skills: 7 SKILL.md files

[0.0.13] - 2025-12-31

Added

• Research Document: Added thinking signature error analysis (docs/research/2026-01-01-thinking-signature-error.md)

Infrastructure

• Build: Stable
• Commands: 17 total
• Agents: 28 total
• Skills: 7 SKILL.md files

[0.0.12] - 2025-12-30

Added

• Spec-Driven Development Workflow: Complete 5-phase workflow (Research → Specify → Plan → Work → Review)
• Visual Workflow Diagrams: Mermaid diagrams in README.md showing the workflow
• Workflow Guide (docs/spec-driven-workflow.md): Comprehensive guide with examples and checklist
• 17 Commands: Added /ai-eng/research as primary workflow command (now 17 total)
• GitHub Methodology Reference: Added link to GitHub's spec-driven development blog post

Changed

• Documentation Cleanup: Removed outdated v0.3.x and v0.4.0 references from all files
• Version Alignment: All documentation now reflects v0.0.x versioning consistently
• Command Structure: Reorganized commands into primary workflow (5) and additional (12)
• Build Optimization: Build time now under 500ms

Fixed

• Plan Archival: Archived 7 outdated plans to archive/plans/
• Version Consistency: Fixed package.json (0.4.0 → 0.0.12), CHANGELOG, TODO.md, IMPLEMENTATION-ROADMAP.md
• Test Suite: All 457 tests passing with no failures

Removed

• Outdated Plans: Removed plans/ directory, consolidated to IMPLEMENTATION-ROADMAP.md

Infrastructure

• Build: 475ms build time
• Tests: 457 pass, 0 fail
• TypeScript: No errors
• Commands: 17 total (5 primary + 12 additional)
• Agents: 28 total
• Skills: 7 files

[0.0.10] - 2025-12-26

Added

• Auto-Installation Plugin Working: Plugin automatically installs commands, agents, and skills when loaded
• 16 Commands: Complete command suite including /plan, /work, /research, /review, /optimize, /deploy, and creation commands
• 29 Agents: Comprehensive agent ecosystem across architecture, development, quality, devops, AI/ML, content, and plugin development
• 13 Skill Files: DevOps, prompting, research, and plugin-development skill packs

Changed

• Documentation Cleanup: Removed outdated v0.3.x and v0.4.0 references
• Version Alignment: All documentation now reflects v0.0.x versioning
• Archived Release Notes: Outdated release notes moved to archive/ directory
• Archived Plans: All outdated plans moved to archive/plans/ directory

Fixed

• Version Consistency: All documentation files updated to reflect v0.0.x versioning
• TODO.md Tracking: Created comprehensive TODO.md for task tracking and documentation cleanup

[0.0.7] - 2025-12-24

Added

• OpenCode Plugin Auto-Installation: Plugin now automatically installs commands, agents, and skills when loaded
• Dual Installation: Works via plugin initialization OR npm postinstall
• Postinstall Script: Automatically detects opencode.jsonc and installs to project directory
• 16 Commands: All commands in ai-eng namespace for OpenCode
• 30 Agents: 5 categories of specialized agents
• 13 Skill Files: Complete skill pack including devops, prompting, research, and plugin-dev
• Usage Examples: Added comprehensive usage documentation in README.md

Fixed

• Installation Path Resolution: Plugin correctly detects if running from dist/ or package root
• Auto-detection: postinstall script finds opencode.jsonc by traversing up directories
• Documentation: Updated command count to 16, skills to 13 files

Changed

• Plugin Implementation: Moved from TypeScript file to pure markdown-based installation
• Configuration: .opencode/opencode.jsonc template includes opencode-skills dependency

[0.0.4] - 2025-12-21

Fixed

• Command registration fixes - all 15 commands now properly available
• NPM publishing workflow implementation
• Removed GitHub Packages configuration for cleaner publishing
• Build artifacts preservation and formatting improvements
• OpenCode agent categories corrected to avoid namespace collision
• Anthropic marketplace pattern implementation with embedded plugin
• Build system fixes for marketplace compliance and mirror validation

Added

• documentation-specialist Agent: Comprehensive technical documentation generation
  • Senior technical documentation specialist with 15+ years experience
  • Generates API docs, user guides, technical specifications, and reference materials
  • Analyzes codebases to create accurate, user-friendly documentation
  • Proactive documentation needs identification for code changes and new features
• Updated agent registry to include all 26 specialized agents
• Fixed swarm integration tests to reflect correct agent count

[0.0.2] - 2025-11-30

Added

• Initial release with core functionality
• Basic agent and command structure
• Documentation system foundation

[0.0.1] - 2025-11-30

Added

• Beta release
• Core system architecture
• Initial command and agent definitions