What's Changed

Codex v4 Agent API Unification

• refactor: Unified all .codex/skills agent calling patterns to Codex v4 API across 31 files in 10 skills
  • wait() → wait_agent() with targets parameter
  • send_input() → assign_task() with structured items array
  • ids → targets parameter naming
  • Skills affected: spec-generator, brainstorm, clean, issue-discover, parallel-dev-cycle, review-cycle, roadmap-with-file, workflow-plan, workflow-tdd-plan, workflow-test-fix-cycle, spec-setup

Features

• feat: Add 4 new UI team skills (interactive-craft, motion-design, ui-polish, visual-a11y) with Impeccable design standards
• feat: Port 4 new UI team skills to Codex v4 format
• feat: Add investigate, security-audit, ship skills (Claude + Codex)
• feat: Enhance skill templates, hooks, CLI routes, and settings UI
• feat: Refactor API calls to use fetchApi for CSRF token handling in Unsplash and Settings pages

Docs

• Add Impeccable acknowledgment to README

What's New

json_builder Tool (Core Feature)

New schema-aware JSON construction and validation tool that replaces manual cat schema + hand-written validation across the entire workflow system.

5 Commands:

• init — Create schema-compliant JSON skeleton from any of 24 registered schemas
• set — Incremental field setting with instant per-field validation (supports path syntax: field.sub, arr[+], arr[0], arr[?key=val])
• validate — Full structural (JSON Schema draft-07 subset) + semantic quality validation
• merge — Merge multiple same-schema JSONs with dedup_by_path strategy (keeps highest relevance)
• info — Compact schema summary replacing raw schema file reads (~100-500 lines → ~20 lines)

Validation Layers:

• Layer 1: Structural — required/type/enum/const/pattern/minLength/minItems/minimum/maximum
• Layer 2: Semantic — generic rationale detection, key_code/topic_relation quality checks, circular dependency detection
• Layer 3: Completeness — all required fields filled with non-empty values

24 Registered Schemas: explore, diagnosis, finding, plan, plan-fix, plan-legacy, fix-legacy, tech, guidelines, issue, queue, review-dim, review-deep, debug-log, discussion, task, solution, verify, discovery-state, conflict, registry, team-tasks, plan-verify

Agent & Skill Updates

All agents and skills updated to use json_builder instead of raw schema file reads:

• Agents: cli-explore-agent, cli-lite-planning-agent, action-planning-agent, issue-plan-agent
• Skills: workflow-plan, workflow-tdd-plan, workflow-test-fix, review-cycle (module + session), workflow-multi-cli-plan
• Commands: workflow/spec/setup

Other Changes

• Workflow session awareness spec added to CLAUDE.md and AGENTS.md
• Removed deprecated prep-cycle.md and prep-plan.md docs

Full Changelog: v7.2.25...v7.2.26

Changes

• Add workflow-research-agent for targeted external research
• Split review responsibilities — code review in lite-execute, convergence review in lite-test-review
• Remove IDAW run and status commands (functionality consolidated into wf-composer)
• Optimize analyze-with-file explore agent triggering for efficiency
• Implement workflow tuning command

Changes

• Add workflow-tune skill for multi-step workflow pipeline optimization
• Enhance workflow-tune with reference document extraction and intent matching
• Enhance spec loading capabilities and add new categories
• Resolve all 92 frontend test failures across 13 test files
• Fix team worker task discovery failures and clean up legacy role-specs

Changes

• Update SKILL.md for analyze-with-file and roadmap-with-file
• Remove analyze-with-file EXECUTE.md
• Remove obsolete tool execution section from documentation

Changes

• Fix: update CodexLens MCP template with AST support defaults

Changes

• Enhance search functionality with quality tiers and scoped indexing
• Enhance CodexLens with quick install feature and embed mode toggle
• Add technical solution recording to analyze-with-file workflow
• Fix dependency graph key, CodexLens env defaults, hook path fixes
• Remove ccw-litellm UI components (dead code)
• Remove codex-lens-v2 from outer git tracking

What's New

New Skill: team-arch-opt (Architecture Optimization)

• New team skill focused on structural architecture improvements: dependency cycles, coupling/cohesion, layering violations, God Classes, dead code
• Isomorphic to team-perf-opt with pipeline: ANALYZE → DESIGN → REFACTOR → VALIDATE + REVIEW
• 6 roles: coordinator, analyzer, designer, refactorer, validator, reviewer
• Supports single/fan-out/independent/auto parallel modes
• 2 subagents: explore (dependency graphs & module boundaries) + discuss (multi-perspective trade-off analysis)

Fixes & Improvements

• fix(analyze-with-file): add session termination boundary and optimize Phase 4 UX
• fix(planning-agent): clarify CLI output as markdown text, relax constraint wording
• fix(planning-agent): relax task parser regex and inject Prior Analysis into CLI context
• fix(planning-agent): remove skip-CLI logic, keep Prior Analysis as context
• fix(planning-agent): enforce blocking CLI calls overriding global default
• fix(planning-agent): add Prior Analysis fast path to skip CLI execution
• refactor(workflow-lite-planex): standardize phase naming and improve documentation

Full Changelog: https://github.com/anthropics/claude-code-workflow/compare/v7.2.0...v7.2.1

🎯 Structured Task Descriptions for Team Workflows

This release introduces a comprehensive structured task description format across all v5 team workflows, providing workers with clearer context and actionable guidance.

✨ Features

team-coordinate-v2: Enhanced Task Analysis & Dispatch

• Structured Task Schema: Tasks now include goal, steps[], key_files[], upstream_artifacts[], success_criteria, and constraints
• Key File Inference: New Step 2.5 in analyze-task.md automatically infers relevant files based on capability type and task keywords
• Structured Dispatch Template: dispatch.md now uses PURPOSE/TASK/CONTEXT/EXPECTED/CONSTRAINTS format

team-lifecycle-v5: Structured Task Descriptions

• Updated Task Templates: All task creation (standard, revision, improvement) now uses structured format
• Clear Worker Context: Workers receive explicit goals, actionable steps, file references, and success criteria
• Improved Traceability: Upstream artifacts and key files are explicitly listed in CONTEXT section

team-skill-designer-v4: Enforced Quality Standards

• Template Specification: Phase 3 now includes detailed task description template for dispatch.md generation
• Quality Verification: Updated quality standards and integration verification to check structured format
• Consistent Generation: All generated v5 skills will use structured task descriptions

📋 Task Description Format

PURPOSE: <what this task achieves> | Success: <measurable completion criteria>
TASK:
  - <step 1: specific action>
  - <step 2: specific action>
  - <step 3: specific action>
CONTEXT:
  - Session: <session-folder>
  - Scope: <scope>
  - Upstream artifacts: <artifact-1.md>, <artifact-2.md>
  - Key files: <file1>, <file2>
  - Shared memory: <session>/shared-memory.json
EXPECTED: <deliverable path> + <quality criteria>
CONSTRAINTS: <scope limits, focus areas>

🔄 Backward Compatibility

• Fully compatible with existing team-worker Phase 1 (reads descriptions as freeform text)
• No schema changes to team-session.json or role-spec templates
• Existing workflows continue to function without modification

📦 Files Changed

• .claude/skills/team-coordinate-v2/roles/coordinator/commands/analyze-task.md
• .claude/skills/team-coordinate-v2/roles/coordinator/commands/dispatch.md
• .claude/skills/team-lifecycle-v5/roles/coordinator/commands/dispatch.md
• .claude/skills_lib/team-skill-designer-v4/phases/03-skill-generation.md
• .claude/skills_lib/team-skill-designer-v4/phases/04-integration-verification.md
• .claude/skills_lib/team-skill-designer-v4/specs/quality-standards.md

🎁 Benefits

• Clearer Worker Context: Workers understand exactly what to do, why, and how to verify success
• Better File Discovery: Key files are inferred and provided upfront, reducing exploration time
• Improved Success Rates: Explicit success criteria and constraints reduce ambiguity
• Consistent Format: All v5 team workflows use the same structured approach

🐛 Bug Fix Release: v7.0.1

This is a critical bug fix release that resolves the frontend deployment issue in v7.0.0.

Fixed

• Frontend Deployment Issue: Fixed "Could not find React frontend directory" error when users install from npm
  • Added ccw/frontend/dist/ to package.json files field
  • Modified react-frontend.ts to detect and use production build (static files)
  • Added static file serving to server.ts with proper MIME type support
  • Updated prepublishOnly script to build frontend before publishing

Technical Details

The issue occurred because:

1. The npm package didn't include the frontend build directory
2. The server tried to start a Vite dev server which requires source files and node_modules
3. Users installing from npm only got the backend code without frontend assets

The fix implements:

• Production Mode: Serves pre-built static files from frontend/dist/
• Development Mode: Falls back to Vite dev server when source files are available
• Automatic Detection: Checks for production build first, then falls back to dev mode
• SPA Routing: Properly handles client-side routing by serving index.html for non-API routes

Installation

npm install -g claude-code-workflow@7.0.1

Verification

After installation, run:

ccw view

The dashboard should now start successfully without the frontend directory error.

Full Changelog: v7.0.0...v7.0.1