All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
3.1.0 - 2026-03-19
/feature-devskill: Guided feature development workflow with 7-phase interactive process — discovery with 95% confidence elicitation, parallel codebase exploration, clarifying questions, multi-perspective architecture design, implementation, quality review, and summary- feature-code agent: New implementation specialist for feature development that reads architecture blueprints from blackboard and creates code following codebase conventions
- Multi-instance agent spawning: Same agent definition can be spawned multiple times with unique names and different focus areas (e.g.,
code-explorer-1,code-explorer-2,code-explorer-3) - Blackboard Protocol sections: All 7 agents now document their blackboard read/write keys in a standardized table
- Config
featureDevsection: Configurable explorer/architect/reviewer instance counts, commit strategy, and PR creation for feature-dev workflow - Feature Architecture Design Mode in architect agent: Designs implementation blueprints with pattern analysis, decisive architecture choices, and phased build sequences
- Mode 4 — Feature Development Review in code-reviewer agent: Focus-area reviews for simplicity/DRY, bugs/correctness, and conventions/abstractions
- Interactive approval gates: Feature-dev includes user decision points at elicitation, clarification, architecture selection, implementation start, and review disposition
- 7-agent architecture: Added feature-code to the roster (was 6 agents)
- code-explorer: Updated to serve both refactoring discovery AND feature-specific exploration with multi-instance support
- architect: Added Bash tool, feature architecture design mode, expanded core responsibilities
- code-reviewer: Added feature development review mode (Mode 4) with focus-area specialization
- refactor-code: Added Bash to allowed-tools
- Config schema version: Bumped to
"3.1"with backward-compatiblefeatureDevkey (defaults applied when missing) - Plugin version: 3.1.0
3.0.0 - 2026-03-18
- Command removed:
commands/refactor.mddeleted — useskills/refactor.mdinstead. The/refactorslash command is replaced by therefactorskill interface. - security-review agent removed: Capabilities merged into
code-reviewer. Any--focus=securityruns now activatecode-reviewerinstead ofsecurity-review. - Config schema version: Bumped to
"2.0". Existingrefactor.config.jsonfiles will be auto-merged with new defaults, but theversionfield should be updated. - Focus flag values changed:
securitynow maps tocode-reviewer(wassecurity-review). Newdiscoveryvalue added.codenow maps toarchitect+code-reviewer(wasarchitectonly).
- code-explorer agent (Phase 0.5): Deep codebase discovery runs first, producing structured codebase maps consumed by all downstream agents
- Blackboard context sharing: Explorer output distributed via Atlatl blackboard with inline fallback when MCP unavailable
--focus=discoveryfocus area for code-explorer-only runsskills/refactor.md— new skill-based interface with full workflow rewrite
- 6-agent architecture: code-explorer, architect, code-reviewer, refactor-test, refactor-code, simplifier (was 5 agents)
- Unified code-reviewer: Merges quality review (confidence-scored, >=80 threshold) with security review (severity-classified, Critical/High = blocking) into single agent
- All agents now use
sonnetmodel (Sonnet 4.6 1M context) — simplifier changed fromopustosonnet - Phase 2 iteration cycle restructured: code review gate (Step 2.E) now precedes simplification (Step 2.F)
- Updated all documentation for 6-agent architecture
agents/security-review.md— capabilities merged intocode-reviewercommands/refactor.md— replaced byskills/refactor.mdcommands/directory — removed (empty after command deletion)
2.2.0 - 2026-03-11
--focus=<area>[,area...]CLI flag to constrain refactoring runs to specific disciplines- Valid focus areas:
security,architecture,simplification,code - Union model for multi-focus:
--focus=security,architecturespawns both discipline agents - Focused runs default to 1 iteration (overridable with
--iterations=N) - Security Posture Score (1--10) rubric in quality scores reference
- Simplification Score (1--10) for simplification-focused runs
- Security-Review Agent section in agent reference
- Focus mode how-to guide (
docs/guides/focus-refactoring.md) - Focus mode troubleshooting entries
- CLI-Only Flags section in configuration reference
- Focus mode architecture explanation
- Agent spawning is now conditional based on
active_agentsderived from--focus - Phase tasks are gated on agent membership (skip steps for inactive agents)
- Commit messages and PR bodies include focus mode label when applicable
- Reports include only scores from active agents
- Updated agent count from "four" to "five" across documentation
- refactor-test and refactor-code always spawn as a safety invariant
2.1.0 - 2026-02-28
- Configuration-driven post-refactor workflow via
.claude/refactor.config.json - Interactive first-run setup wizard with AskUserQuestion prompts
- Commit strategies: none, per-iteration, single-final
- Optional PR creation (draft or ready-for-review) after refactoring
- Report publishing to GitHub Issues or GitHub Discussions
- Cross-referencing between PRs and published reports
- Non-blocking error handling for all GitHub operations
- Self-contained git operations (no external plugin dependencies)
- Diataxis-structured documentation (tutorial, how-to guides, reference, explanation)
- Social preview images (light + dark, illustrated SVG style)
- README infographic with theme-switching
<picture>element
- Replaced external
commit-commands:commitskill references with inline git sequences - Replaced external
commit-commands:commit-push-prskill references with inline gh CLI sequences - Replaced
git add -Awithgit add -ufor safer staging - Updated config schema version from 1.0 to 1.1
- Added
iterationsandreportRepositoryfields to config schema
2.0.0 - 2026-02-01
- Swarm orchestration (TeamCreate, TaskCreate/TaskUpdate, SendMessage)
- New simplifier agent (opus model) for code clarity passes
- Parallel execution in Phase 1 (foundation) and Phase 3 (final assessment)
- Code simplification step after each iteration cycle
- 4-phase workflow replacing 7-step sequential process
- Sequential execution model
1.0.0 - 2026-01-01
- Initial release with sequential 7-step workflow
- Three agents: architect, refactor-test, refactor-code