Changelog
A complete history of features, fixes, and changes in H·AI·K·U (formerly AI-DLC).
Fixed
- Reverted haiku rebrand to previous state
Added
- Design provider support — reference design artifacts from Figma, Canva, Excalidraw, OpenPencil, Pencil, and Penpot; the plugin auto-detects and integrates them into your visual review workflows
- Post-integrate knowledge refresh — knowledge context automatically updates after integration runs
- Native design provider link resolution — handles both web URLs and native provider links for seamless artifact reference
Changed
- Designer hat now leverages design provider artifacts in visual reviews and wireframe elaboration for context-aware feedback
Added
- Subscribe to changelog and blog updates via RSS, Atom, and JSON feeds
- Release-notes skill for accessing release information
Fixed
- Feed generation now properly escapes URLs in XML and formats Atom dates correctly
Changed
- Reorganized internal subskills to nest under their parent skill directories for clearer plugin structure.
Fixed
- Corrected stale command references and paths throughout skill definitions and artifacts to ensure proper subskill invocation.
Fixed
- Agent teams created during a unit are now properly cleaned up when the unit completes.
Changed
- Changelog generation is now AI-synthesized for consistency across release notes.
Fixed
- Reverted Visual Review full integration pending further development.
Added
- /ai-dlc:review skill for pre-delivery code review with automated fix loop, integrated into quick mode
Changed
- Automatic setup now runs if settings.yml is missing
- Quality gates handling improved with clearer formatting and support for 5+ gates
- Elaboration now less verbose with simplified setup concerns
- PR terminology standardized across skills and documentation
Fixed
- Plugin installation is now more reliable and no longer requires npm
- CPU architecture is automatically detected for cross-platform binary dependency support
Added
- Knowledge synthesis and design direction system with archetype library and visual picker for elaboration bootstrapping
- Three new elaboration phases: 2.3 (Knowledge Bootstrap), 2.75 (Design Direction), 7.5 (Adversarial Review), plus autonomous elaboration mode for /autopilot
- /adopt skill to reverse-engineer existing code into AI-DLC intent artifacts
- Dynamic pass definitions allowing custom passes in construction pipeline
- Quality gates with visual review interface, intent dashboard, and mermaid-enabled MCP server
- BDD workflow with acceptance-test-writer hat for behavior-driven elaboration
Removed
- workflow_mode setting — /autopilot is now autonomous by definition
No notable changes.
Added
- Visual-review intent to structure collaborative design and UX review workflows
- Sync rules documenting consistency requirements between paper, plugin, and website components
Fixed
- Intent status is now validated before delivery to catch configuration issues early.
Fixed
- Intent status now correctly updates when Teams path completes
Removed
- Standalone HTML guides (content now available on website)
Fixed
- Planner hat no longer incorrectly enters plan mode, strategy options are properly ordered, and intent strategy no longer prompts for auto-merge confirmation.
Added
- Story-driven home page and plugin lifecycle guides provide intuitive introduction to the AI-DLC methodology and workflow.
Fixed
- Intent strategy is now the default workflow, eliminating the auto-merge confirmation prompt.
Added
- Passes are now configurable, letting you customize which disciplinary lenses (design, product, dev) iterate in your workflows.
Changed
- Construction Phase is now called Execution Phase.
Fixed
- Plugin hooks no longer depend on external han library, reducing potential failure points and improving reliability.
Added
- Harness framework documentation in the methodology paper and website, explaining how the plugin execution model works.
Fixed
- Removed outdated han-plugin references from skill documentation.
Fixed
- Advance skill now properly evaluates completion gates against the actual plan file
- Resume skill correctly parses YAML frontmatter in unit specifications
- Haiku utility can now be called standalone with automatic state dependency resolution
Added
- Infrastructure and operations as a workflow discipline: Elaborate now guides you through infrastructure questions and auto-creates operations units alongside feature work
- Operations construction phase that builds and validates deployment and infrastructure configurations during the execution flow
- Operational readiness reviews that validate infrastructure and deployment completeness before shipping
- Stack configuration schema with quality gates for defining infrastructure patterns and architectural standards
Fixed
- Fixed visual-fidelity issues identified in code review
Added
- Completion criteria checkboxes automatically check off when a unit or intent is completed, keeping your workflow state synchronized with task progress.
Fixed
- Interactive workflow mode now has write permissions enabled.
Fixed
- Code review workflow now has the necessary permissions to write and apply changes when reviewing code.
Added
- Visual fidelity review agent with gate detection and comparison orchestration for automated design quality assessment, including backpressure documentation for integration workflows
- Design reference resolver that intelligently locates design system components using a 3-level priority hierarchy
- Pluggable screenshot capture infrastructure enabling flexible image acquisition across different design tools
Fixed
- Prevent stderr/JSON mixing in design reference resolution calls
Fixed
- Intent and unit completion statuses are now reliably set when work finishes.
Changed
- Plugin no longer depends on han CLI; core state management and hook utilities are now built into the plugin via foundation libraries
- Simplified iteration.json state schema with improved phase validation
Added
- Enforce quality checks with hard gates that must pass before unit completion.
- Automatically delegate work to specialized subagents using configurable hooks.
- Generate DOT flowcharts to visualize intent and unit workflows.
- Validate execution context with preflight checks before running workflows.
No notable changes.
Added
- Code reviews now support multi-judge critique debate for high-stakes decisions.
Fixed
- Fixed CI pipeline conflict resolution to properly handle push events.
Fixed
- Context:fork skill invocations during elaboration now use the Agent tool for improved execution reliability
Added
- Schema drift detection for database changes during code review
Added
- Use /autopilot skill to run complete autonomous feature lifecycles from elaboration through delivery.
Changed
- /execute is now the primary execution skill; /construct is deprecated.
Added
- Reviewers can now use structured completion markers to mark deterministic review outcomes with confidence-based feedback.
Added
- Builder includes a node repair operator and structured completion markers for improved construction reliability and state tracking.
Fixed
- Improved conflict resolution in continuous integration builds using matrix strategy.
Added
- The planner now analyzes git history to provide more informed planning recommendations.
Fixed
- Shell scripts are now executable after installation, eliminating permission errors.
Added
- Workflow mode selection and granularity tuning options to customize AI-DLC iteration behavior.
- Documentation for integrating Reflect into the Compound skill workflow.
Added
- DOT flowchart guidelines for visualizing process authority during the construction phase.
Added
- Model profiles for cost-optimized hat routing with improved configuration resolution
Added
- Planning process now retrieves relevant learnings before generating plans, with improved search capabilities
Fixed
- Restored Structured Completion Marker section that was accidentally removed
Fixed
- Restored missing Two-Stage Review, conflict resolution, CoVe, and Specialized Pre-Delivery Reviews sections in reviewer hat
Added
- Parallel review perspectives for multi-file units enable concurrent reviews from multiple disciplinary angles
Added
- Three-level artifact checking in reviews validates content, reasoning, and structural integrity
- Goal-backward verification ensures work aligns with original intent throughout development
- Anti-rationalization tables and red flags across all hats identify biased reasoning patterns
- Confidence scoring system enables granular quality assessment during review cycles
- Enhanced anti-pattern detection and specialized review modes for deeper code analysis
Fixed
- Conflict resolution in automated workflows now requests manual review instead of taking automatic action, improving reliability when handling merge conflicts.
Changed
- Documentation terminology updated throughout to reflect current plugin naming, removing references to legacy terms.
Fixed
- Conflict resolver polling now operates correctly during workflow execution.
Added
- Customize elaboration phases for your project using .ai-dlc/ELABORATION.md configuration files
- Configure external rule files using a data-driven pattern
No notable changes.
Changed
- Reviewer hat now validates spec compliance first (including test coverage requirements), then assesses code quality in a separate stage.
Added
- Visual brainstorming companion guidance during the elaboration phase
Fixed
- Brainstorm storage path to prevent collisions with mockups directory
No notable changes.
Added
- Context budget monitoring warns when context usage approaches limits at 35% and 25% remaining, helping you manage token usage during long sessions.
Changed
- Builder now reduces token consumption by approximately 35% by extracting reference materials to a companion file.
Changed
- Reduced token consumption during code reviews by extracting reference material to a companion file.
Added
- Structured completion markers explicitly control when the workflow transitions between hats, replacing implicit transition logic with deterministic gating.
Added
- Code reviews now include confidence-scored findings and automated anti-pattern detection.
Changed
- Session token consumption is reduced through optimized hook lazy loading
Added
- Capture learnings from the current session as structured solution files with the new `/compound` skill.
Added
- Use anti-patterns guidance when writing completion criteria to avoid common mistakes that over-constrain or inappropriately restrict agent output.
Added
- Added `/pressure-testing` skill for test-driven development with quality gate validation.
Added
- You can now disable model invocation in the reflect, resume, and operate skills to prevent unintended model calls during execution.
Added
- Execute multiple units within an intent in parallel waves, automatically coordinating dependencies and sequencing units for optimal throughput.
Added
- Sessions now resume with structured handoff to maintain context and progress across bolts.
No notable changes.
Changed
- Review operations now consume fewer tokens due to optimized subagent context scoping
Added
- Discovery phase now supports brownfield codebase mapping to help you understand existing code architecture when elaborating new intents.
Added
- Aggregate learnings from compound work into structured reflection outputs for project documentation and review.
Added
- Workflows now enforce quality gates at hard-gate synchronization points during critical phase transitions.
Added
- Elaboration now includes automated specification review before advancing to construction.
Added
- Builder now requires verification of completion criteria before marking units complete.
Added
- Use the `/seed` skill to capture forward-looking ideas and define conditions for when they should be revisited.
Changed
- Reviewer now uses specialized review agents to provide comprehensive pre-delivery checks instead of a static checklist
Added
- Git strategy questions now show your actual default branch name, and you can configure which remote to use for detection.
Added
- Claude Code Review workflow for conducting code reviews directly within Claude Code.
Changed
- Improved Claude PR Assistant workflow with enhanced pull request management capabilities.
No notable changes.
Added
- Plugin state now persists to disk with lockfile protection, enabling safe multi-session workflows.
Added
- Safely execute untrusted or generated prompts with built-in prompt injection guards
- Configure workflow enforcement hooks to control tool execution and maintain process discipline
Added
- Use `/quick` mode to skip full elaboration for trivial tasks.
Added
- Agent teams can now use a parallel commit strategy with no-verify enabled to speed up concurrent builds.
Added
- Specialized review agents organized by domain, enabling more targeted feedback based on codebase expertise.
Added
- You can now use `/backlog` to capture and organize parking lot ideas during development.
Added
- Generate improvement suggestions with adversarial review using the new `/ideate` skill.
Added
- Run parallel research agents to deepen plans and explore multiple angles simultaneously
Added
- Spec flow analysis during elaboration helps identify specification gaps before implementation.
Added
- Customize review agent configuration on a per-project basis to align code review behavior with your team's standards.
Added
- Reviewer now uses chain-of-verification (CoVe) methodology to provide evidence-based code reviews grounded in systematic analysis.
Added
- Search for learning materials in the planner and get results ranked by relevance using multiple signals to quickly find applicable guidance.
Added
- Builder automatically detects version incompatibilities and provides rollback guidance during builds.
Added
- The planner now uses rule-based filtering to automatically select the best approach for your task.
Added
- Configuration documentation now explains the precedence pattern for master and override settings, allowing you to correctly configure layered settings.
Added
- Units now include a last_updated timestamp in their frontmatter, allowing you to track when each unit was last modified.
Fixed
- Unit and intent status changes are now committed to git automatically, ensuring all workflow state transitions are tracked in version control.
Changed
- Strategy options documentation now explains each choice from the user's perspective, making it clearer when to use each strategy.
Added
- Automatic cleanup of temporary git worktrees at completion milestones, eliminating manual cleanup steps after intent completion.
Added
- Design-specific unit template sections to guide structured documentation of design requirements during intent elaboration.
Changed
- `/followup` now creates linked iteration intents to maintain continuity across intent iterations.
Added
- Wireframes now include design units during the elaboration phase for more complete design specifications.
Added
- Receive design-focused guidance when defining success criteria during the elaborate phase.
Added
- Design discipline now automatically routes to a design-focused workflow.
Added
- Plugin now automatically detects and removes worktrees once their changes are merged.
Added
- Workflow execution events are now reported to OpenTelemetry, enabling integration with observability platforms and monitoring tools.
Changed
- Documentation rebrand with no user-facing changes
Added
- Test-driven development workflow for building features from tests first
- Learnings from completed intents are automatically loaded into new workflows
Changed
- Integration workflows now run autonomously without manual intervention
Removed
- Operator and Reflector hats and their associated workflows have been removed from the available workflow options.
Fixed
- Plugin can now be properly discovered and installed via the Claude Code marketplace
Fixed
- Unit strategy automations no longer ask for auto-merge confirmation during setup.
Changed
- Simplified plugin source configuration.
Changed
- Restructured plugin directory and marketplace configuration
Added
- `/reflect` skill for reflection phase—review and learn from completed work
- `/operate` skill for operation phase—maintain and iterate on deployed systems
- AI-DLC integrates with HAIKU software development methodology
Changed
- Construction phase renamed to Execution throughout plugin and documentation
Added
- Plugin can be downloaded as a zip archive
Changed
- Marketplace now uses path-based source configuration
Changed
- H·AI·K·U is now discoverable in the plugin marketplace with full description
No notable changes.
Fixed
- Plugin now correctly resolves worktree paths relative to the main repository root instead of the worktree directory, preventing path resolution errors when using git worktrees.
Fixed
- Review pull requests created during elaboration no longer close linked issues when merged.
Fixed
- Unit-based change strategy no longer prompts for delivery confirmation, streamlining the workflow.
Fixed
- Elaboration review now displays complete unit content without truncation.
Fixed
- Discovery skill now enforces strict alignment rules for ASCII wireframes, improving consistency and accuracy during domain exploration.
Added
- Elaborate phases can now be run independently as fork subagent skills, giving you granular control over which phases to execute.
Fixed
- Worktree directories are now automatically added to gitignore, preventing temporary work from being accidentally committed.
Fixed
- Elaboration workflow is now unified, removing the complexity of managing separate elaborator agent and elaboration-start skill components.
Added
- Changelog page on the website for viewing release history
Changed
- Elaborate workflow now orchestrates multiple agents for better intent specification
Added
- Detect greenfield projects and automatically generate UI mockups for rapid design exploration.
Fixed
- Changelog generation now includes only commits since the previous version for more accurate release notes
Fixed
- Intent discovery now creates a worktree before running, preventing temporary artifacts from appearing on the main branch.
Added
- Agents can now invoke elaborate, resume, and refine skills, enabling automated AI-DLC workflows.
Added
- Discovery scratchpad automatically captures elaboration findings to `discovery.md` during domain discovery, reducing context window pressure while keeping detailed findings accessible to subagents and the intent worktree
- Parallel design analysis via dedicated subagents during elaboration for faster, context-efficient Figma and design file processing
- Hybrid per-unit change strategy support, allowing mixed strategies within a single intent—foundational units can use direct-to-main while remaining units merge to the intent branch
Added
- Per-unit workflows enable different workflow configurations for each unit, with built-in support for design discipline phases.
Added
- Handle design assets directly in your workflows for easier design integration
- Automatically match and apply colors from design specifications
- Leverage design annotations to guide code generation and specifications
Added
- Handoff now supports cowork-aware modes with options to share work as a local folder or zip file for easier collaboration.
Changed
- Slash commands `/reset`, `/refine`, `/setup`, and `/resume` are now blocked in cowork mode to prevent conflicting simultaneous operations.
Changed
- The `/construct` command is now unavailable in cowork mode to prevent mode conflicts during collaborative work.
Added
- Cowork mode now automatically detects multi-agent collaboration environments, improving coordination when multiple Claude Code instances work together.
- Exploration tasks now delegate to specialized Explore subagents for faster codebase discovery and analysis.
Added
- Wireframe generation phase for visual design exploration and iteration in development workflows
Changed
- Worktrees are now automatically created within project directories for better isolation and organization
Fixed
- Ticket descriptions now display with improved formatting and structure for better readability.
Changed
- Reference skills are now internal-only and cannot be invoked directly
No notable changes.
Added
- Execute specific units without running the entire intent
Changed
- Change strategies now provide more options for managing workflow iterations
Removed
- Bolt change strategy has been removed; use the intent strategy instead
Added
- Completion announcements signal when tasks meet their completion criteria.
- Risk descriptions let you document potential issues identified during elaboration.
- Iteration caps set a maximum limit on how many times a bolt can cycle.
- Explicit bolt terminology clarifies the iteration cycle concept throughout the plugin.
Changed
- Simplified skill definitions and hook configurations by removing mode references
Removed
- Mode selection from elaboration and construction phases.
Fixed
- Iteration state is now correctly initialized during construction instead of elaboration, ensuring proper workflow sequencing from project inception through delivery.
Added
- **Integrator hat**: Validate cross-cutting concerns after merging all units within an intent.
- **/refine skill**: Refine intent or unit specifications mid-execution without losing progress.
- **Non-functional requirement guidance**: Enhanced prompts to help you specify and validate non-functional requirements and cross-cutting concerns.
- **Delivery phase guidance**: Improved prompts for planning and executing the delivery phase.
Added
- New `/setup` skill automates initial project configuration.
- Elaboration now enforces ticket creation for all work before execution can begin.
Fixed
- Testing requirements are now mandatory for all intents and cannot be configured per-intent.
Fixed
- Subagents now load context state from the correct branch when executing work.
Fixed
- Intent branches are now prefixed with `ai-dlc/` to prevent naming conflicts and improve branch organization.
Fixed
- Construction workflows now skip the elaborator phase and intent discovery has been improved.
Fixed
- Plugin installation now uses Claude Code's native `/plugin` CLI for streamlined setup.
Added
- Integrate external providers to customize how agents access memory, data, and external services
- Enable coworking mode for team-based and collaborative AI-DLC workflows
- Instructions now merge hierarchically across global, project, and local levels so local customizations take precedence
Added
- Configure memory providers to customize where and how AI-DLC state is stored.
- Collaborate with team members through cowork support for multi-user development workflows.
Changed
- Plugin structure reorganized for improved modularity and maintainability.
Changed
- Intent configuration is now consolidated into intent.md frontmatter, eliminating the need for a separate intent.yaml file.
Fixed
- Session retrospectives now correctly handle branch ordering and workflow transitions
- Team mode hats now function properly in multi-agent workflows
- Merge strategy now executes as configured
Added
- Domain discovery during elaboration helps you systematically explore your problem space before constructing solutions
- Specification validation during elaboration catches design issues early in the inception phase
- Deep research capabilities help you gather comprehensive context about relevant technologies and approaches
No notable changes.
Added
- Agent Teams with intent-level modes and dynamic hat discovery
- Interactive workflow visualizer on documentation website
- Website responsive layout and dark mode
Changed
- Construction workflows now use Agent Teams by default
Fixed
- Session startup performance optimized