v2.1.0

Highlights

• Dual-output review agents — All 9 review/checker agents (code-quality, security, performance, test-quality, architecture-alignment, spec-alignment, design-consistency, housekeeper, validation-fixer) now write findings to a file and stream to stdout, so the orchestrator can read them without re-spawning.
• Pane-based project state — Project selection is now keyed off terminal pane TTY rather than session id, so concurrent Claude sessions running in different panes of the same repo no longer clobber each other.
• Validation-loop overhaul — Budget tracking, clearer fix-then-rerun flow, and tighter convergence semantics.
• execute-all-tasks overhaul — Stricter task-dependency handling and worktree lifecycle.

v2.0.0

What's New

Major Features

Monorepo Support

• New repo-setup skill detects single-project vs monorepo structure and creates .groundwork.yml
• New project-selector skill for switching between projects in a monorepo
• Monorepo-aware paths and template variables across skills, hooks, and commands
• Cross-platform monorepo command support

Agent Teams & Swarming

• New swarm-debugging skill: parallel hypothesis testing with adversarial agent teams
• New swarm-architecture-design skill: parallel advocate agents build cases for competing technology options
• New execute-all-tasks-swarming skill: each task runs in an isolated teammate session that can spawn its own subagents
• Aligns with the constraint that subagents cannot nest — teams provide context isolation instead

PR Reviewing

• New pr-reviewing skill runs multi-agent verification against PR diffs and posts inline/general comments to GitHub
• New review-pr command

Design System

• New design-system skill: foundations, brand identity, and UX patterns in one workflow
• sync-design-system skill for end-of-session token/pattern sync
• Improved UX design practices throughout

Auto-Splitting Large Specs

• Product specifications and architecture docs auto-split into directory-based format when they grow too long
• New split-specifications and split-architecture-doc skills with corresponding commands
• New impact-analysis skill traces blast radius of requirement changes across split specs

Validation & Quality

• New test-quality-reviewer agent for structural correctness, coverage completeness, and testing best practices
• New validation-fixer shared agent: applies TDD for behavioral fixes and direct changes for cosmetic ones
• Reinforced validation loop with autonomous fix-and-retry until all agents approve
• Stale-test fix: hooks.json correctly references PostToolUse, not PreToolUse

Workflow Improvements

• Tasks now organized into folders and individual files instead of one monolithic list
• task-executor agent uses inherit model setting (no longer forces Opus)
• TDD skill simplified — no longer requires specific models
• Compact pre-validation and between tasks to keep context bounded
• Better debugging journal detail and hypothesis tracking
• Replaced TTY-based session IDs with Claude's native session_id
• Section numbering, cross-references, and skill workflows refactored for clarity

Install Targets

• Added Pi coding agent as install target for install-skills.sh

Breaking Changes

• Tasks directory structure changed: tasks are now split into folders and individual files
• Architecture docs and product specs use directory-based format when split
• Removed unused skills and commands; several skills renamed or refactored
• Replaced Task( with Agent( in all skill code examples (the actual tool name)
• Personal skills in ~/.claude/skills/ still shadow plugin skills — use the groundwork: prefix to force the plugin version

Documentation

• Expanded README with prefix conventions and removed redundant skills section
• Updated agent and skill documentation throughout
• Recommended models and effort levels documented per skill

Stats

• 30 skills, 16 agents, 25 commands

v1.2.0

What's New

Cross-Platform Installer

• New install-skills.sh script for installing Groundwork skills on non-Claude platforms (Cursor, Windsurf, Cline, etc.)
• Configurable via install-config.txt

Command-to-Skill Refactoring

• 5 commands refactored into proper skills: check, help, execute-all-tasks, skills, split-spec
• Commands now delegate to skills via Skill tool invocation, improving reusability

Other Changes

• Removed Windows hook runner (run-hook.cmd)
• Expanded README with installation and usage documentation

Stats

• 24 skills, 14 agents, 18 commands

v1.1.0

What's New

Design System Skill Enhancements

• Atmosphere Comparison (Step 7b) — Side-by-side visual comparison of design atmosphere variations
• Pattern Showcase (Step 7c) — Component pattern gallery for design system review
• UX Preview Generation (Step 8b) — Generate interactive UX previews from design tokens
• Restructured workflow — Step 7 reorganized into focused sub-steps for better flow

Sync Design System Updates

• UX Preview Regeneration — New step 6 ensures UX previews stay in sync when design tokens change

Full Changelog: v1.0.4...v1.1.0

v1.0.4

Changes

• Smarter validation loops: Skip irrelevant agents when their review subjects don't exist (e.g., design-consistency-checker skipped when no design system is present)
• Selective re-runs: On retry, only re-launch agents that returned request-changes instead of all agents, with domain spillover detection for cross-cutting fixes
• Clearer next-task handoff: Improved instructions for delegating to execute-task with explicit task-id format

v1.0.3

What's Changed

• execute-task: Improved plan mode handling with clearer execution context, merge mode, and skip-to-step-7 support
• next-task: Added guardrails to prevent premature planning or building before full workflow execution
• Version bump to 1.0.3

v1.0.2

Full Changelog: v1.0.0...v1.0.2

v1.0.0

Groundwork v1.0.0

Initial release of the Groundwork Claude Code plugin — a comprehensive skills library for structured development workflows.

Highlights

• 19 skills covering planning, design, TDD, debugging, and synchronization
• 18 commands for user-invoked workflows
• 7 agents for specialized sub-tasks (code review, security, performance, etc.)
• Session hooks for automatic context loading and state preservation

Skills Categories

• Product & Architecture: product-design, architecture, check-alignment, sync-specs, sync-architecture
• Implementation: tasks, execute-task, next-task, implement-feature, test-driven-development, build-unplanned-feature
• Quality: validation-loop, task-validation-loop, debugging
• Design: design-system, sync-design-system, frontend-design
• Workflow: using-groundwork, understanding-feature-requests, use-git-worktree