Release: v0.10.8

Design Document Quality Improvements

Added six enhancements to strengthen design document rigor across the creation and review workflow.

New Required Sections in Design Template

• Applicable Standards — Classify project standards as explicit (documented) or implicit (observed), requiring confirmation for implicit ones
• Code Inspection Evidence — Record all inspected files/functions with their relevance during investigation
• Data Representation Decision — Evaluate reuse-vs-new using semantic fit, responsibility fit, lifecycle fit, and boundary cost criteria
• Field Propagation Map — Track field status (preserved / transformed / dropped) at each component boundary

Technical Designer Updates

• Standards Identification Gate — Must identify and classify project standards before any investigation begins
• New checklist items enforce recording of all new required sections

Document Reviewer Updates

• Gate-based review structure — Step 3 now uses Gate 0 (structural existence) before Gate 1 (quality assessment); Gate 0 failure blocks quality review
• Rationale verification — Design decisions must reference identified standards or existing patterns
• Code inspection evidence review — Verify inspected files are relevant and flag missing key files
• Updated review criteria: Gate 0 pass is required for both "Approved" and "Approved with Conditions"

Release: v0.10.7

What's New

New Command: /update-doc

Added a new command to update existing design documents (Design Doc / PRD / ADR) with a proper review workflow.

Features:

• Identifies and selects target documents from docs/design/, docs/prd/, or docs/adr/
• Clarifies changes with user before updating
• Updates via technical-designer or prd-creator in update mode
• Validates through document-reviewer and design-sync
• Includes ADR update guidance (minor vs major changes)

Available in both plugins:

• dev-workflows (backend)
• dev-workflows-frontend (frontend)

Release: v0.10.6

New

• Add metronome as an external plugin to the marketplace
  • During repetitive tasks, Claude may switch to bulk bash operations under the guise of "working efficiently," potentially breaking files. Metronome detects that behavior and intercepts execution, nudging Claude to slow down and proceed step by step.
  • Install: /plugin install metronome@claude-code-workflows

Changed

• Update marketplace description to reflect broader scope (plugins, tools, and safety guardrails)
• Update README with External Plugins section

Release: v0.10.5

Highlights

This release improves command execution reliability by ensuring all steps are properly tracked in TodoWrite and delegated to subagents.

Changes

Command Structure Improvements

• First Action directive: Added First Action: Register Steps 0-N to TodoWrite before any execution to orchestrator commands
• Step 0 inclusion: Step 0 (skill execution) is now part of the main flow and tracked in TodoWrite
• Unified step notation: Standardized all commands to use ### Step N: format
• Explicit fix steps: Added dedicated steps for applying review fixes (ensures delegation to task-executor)

Removed Deprecated Patterns

• Removed Think harder/deeply expressions
• Removed @agent-name notation (replaced with Task tool invocation format)
• Removed indirect skill references (see documentation-criteria skill)

review.md Simplification

• Removed redundant rule-advisor step (code-reviewer already identifies fixes)
• Simplified conditional flow with single skip condition at Step 4
• Reduced steps from 10 to 9

add-integration-tests.md Enhancements

• Added Core Identity and Why Delegate sections
• Added Step 6: Apply Review Fixes for explicit fix delegation
• Embedded task file template directly in command

Fixed Issues

• Step 0 being skipped because it was outside TodoWrite registration scope
• Orchestrator performing direct edits instead of delegating to task-executor
• Subagent invocation format inconsistencies across commands

Release: v0.10.4

Improvements

• add-integration-tests command: Improved command format with explicit Task tool invocation blocks
  • Added **Role**: Orchestrator declaration
  • Added **Task tool invocation**: blocks with subagent_type and prompt for each sub-agent call
  • Added **Expected output**: for each step
  • Added **Flow control**: for review step branching
  • Aligned format with diagnose and reverse-engineer commands

Release: v0.10.3

Orchestration Workflow Improvements

Step Count Verification

Added explicit step counts to workflow flow headings (Large: 11 steps, Medium: 7 steps, Small: 2 steps). This enables orchestrators to self-verify that all required steps are executed without skipping.

Structured JSON Output for requirement-analyzer

Converted requirement-analyzer output to mandatory JSON format with:

• scopeDependencies: Questions that may change scale determination based on user answers
• adrRequired / adrReason: ADR necessity with explicit reasoning
• Removed redundant fields (PRD/Design Doc requirements are derivable from scale)

Improved Scale Determination

Added Grep tool to requirement-analyzer for more accurate file count estimation during scale assessment.

Explicit Stop Point Behavior

Added explicit requirement to use AskUserQuestion at all Stop points, ensuring proper user interaction during workflow execution.

Streamlined Response Specification

Updated Structured Response Specification to include only decision-relevant fields for each subagent, reducing context overhead.

Purpose-Driven First Action Rule

Clarified First Action Rule with purpose-driven description: pass requirements directly to requirement-analyzer for accurate analysis before workflow determination.

Release: v0.10.2

Bug Fixes

• Stabilize agent JSON output: Add explicit "JSON format is mandatory" directive to Output Format sections in code-verifier, scope-discoverer, investigator, and verifier agents. This prevents agents from returning prose summaries instead of structured JSON, ensuring downstream workflows can reliably parse agent output.

Affected Workflows

• reverse-engineer: code-verifier and scope-discoverer output
• diagnose: investigator and verifier output

Release: v0.10.1

Fixes

• Remove command reference from subagents-orchestration-guide skill to maintain single responsibility principle
  • Skills should not reference specific commands
  • Command files (commands/add-integration-tests.md) are self-contained

Release: v0.10.0

New Features

Backend

• /add-integration-tests - Add integration/E2E tests to existing codebase using Design Doc
• integration-test-reviewer - Now properly triggered in 4-step cycle when *.int.test.ts or *.e2e.test.ts are added

Frontend

• /front-review - Frontend-specific code review command (uses task-executor-frontend, quality-fixer-frontend)
• /front-reverse-design - Generate frontend Design Docs from existing PRD

Improvements

• Task template commonization - Shared task-template.md in documentation-criteria skill
• Mermaid diagram update - Autonomous execution flow now shows integration-test-reviewer branch
• README updates - Improved command documentation for both plugins

Breaking Changes (Frontend)

Before	After
/review	/front-review
/reverse-engineer	Removed (use /front-reverse-design with existing PRD)

Agent Updates

Plugin	Before	After
Backend	17 agents	18 agents (+integration-test-reviewer)
Frontend	14 agents	16 agents (+design-sync, +rule-advisor)

Release: v0.9.7

Command Execution Accuracy Improvements

This release focuses on improving command execution accuracy by addressing structural issues that caused orchestrators to deviate from intended behavior.

Explicit Delegation Mapping

Added clear work-to-agent mapping in Execution Method sections:

**Execution Method**:
- Investigation → performed by investigator
- Verification → performed by verifier
- Solution derivation → performed by solver

This prevents orchestrators from performing tasks directly instead of delegating to sub-agents.

Anti-Pattern Removal

Pink Elephant Problem: Replaced "NOT included" and "Avoid" sections with positive "Responsibility Boundary" statements. Telling LLMs what NOT to do can inadvertently focus attention on prohibited actions.

Command Coupling: Removed cross-references between commands (e.g., "Use /plan after approval") to reduce unnecessary dependencies.

Decorative Noise: Removed decorative emojis from section headers to reduce context consumption.

Structural Additions

• Added Orchestrator Definition to commands missing explicit delegation (front-build, front-design, front-plan, review)
• Added Workflow Overview diagrams for complex flows (design)
• Added Completion Criteria checklists (design)

Files Changed

• diagnose.md, design.md, plan.md
• front-build.md, front-design.md, front-plan.md
• review.md, build.md, implement.md