Changelog

The website keeps recent public release highlights. For the full engineering history, see the full GitHub changelog

v2.4.0

Major

This was not a light polish release. It was a multi-day product-shaping overhaul across the host model, onboarding lifecycle, recovery chain, UI coaching, and release gates.

  • -The public product surface is now fully host-first: the terminal keeps only super-dev / super-dev update / super-dev uninstall, while baseline, research, the three core docs, approval gates, spec, implementation, and delivery all happen inside the host
  • -Existing projects now continue on a more complete path: establish the current baseline first, then move through research, the three core docs, approval, frontend preview, backend, quality, and delivery
  • -Host onboarding was rebuilt around a unified host matrix (CLI / IDE / desktop assistants), with Droid CLI, Kimi Code, Qwen Code, Trae SOLO, and Trae SOLOCN treated as first-class hosts; OpenClaw has been removed from the current support surface
  • -Onboarding is now project-first by default: connect the current repo first, then expose the host first prompt, resume path, and post-onboard self-check so install is not mistaken for true readiness
  • -Dual-mode readiness is now explicit: users can see whether standard product work and SEEAI competition work are actually ready to start instead of just seeing that files were written
  • -super-dev-core was purged from current code, templates, skills, and active test contracts. The product name is now consistently super-dev, and rules default to project-level surfaces instead of silently writing into system-level AGENTS.md
  • -The UI design chain was upgraded: the UI contract now freezes art-direction candidates, a visual philosophy, anti-AI-slop guardrails, a five-dimension critique rubric, and tweak categories, while Framework focus and the screenshot-grade UI visual gate now participate in quality / proof-pack / release-readiness
  • -Release gates are green again: pytest 2576 passed, 2 skipped; preflight --allow-dirty passes; the core release gates, knowledge gates, delivery smoke, host compatibility, and build/twine-check are all passing

Product shape reset

  • The terminal was reduced to install/update/uninstall while the real workflow moved back into the host.
  • Public messaging was pulled away from “toolbox / scaffolding” language and back toward a host coaching system.
  • The first-layer user path now centers on /super-dev, /super-dev-seeai, continue current workflow, and what next.

Host matrix and onboarding

  • The unified host matrix now spans CLI, IDE, and desktop assistant hosts, with Droid CLI, Kimi Code, Qwen Code, Trae SOLO, and Trae SOLOCN elevated to first-class status.
  • Project-first onboarding is now the default, while user-level surfaces became an explicit opt-in.
  • Each host now exposes first prompts, resume guidance, repair playbooks, official workflow checks, and post-onboard self-checks.

Existing-project flow and recovery

  • baseline -> docs -> docs_confirm -> spec -> frontend -> preview_confirm -> backend -> quality -> delivery became the formal path.
  • resume / next / continue / doctor / validate now explain the current blocker, the next step, and why the workflow cannot skip ahead.
  • Onboard smoke guides, uninstall dry runs, and cleanup reports are now formal lifecycle outputs.

UI quality and premium delivery

  • The UI contract now captures art direction, brand signals, proof composition, component craft, layout tension, and anti-AI-slop guardrails.
  • The screenshot-grade UI gate now participates in quality, proof-pack, and release readiness.
  • Framework Coaching Focus now reaches the installer, runtime reports, smoke guides, and delivery summaries.

Release and engineering quality

  • Preflight, knowledge gates, delivery smoke, host compatibility, build, and twine-check all returned to green.
  • Public docs, the website, the installer, and changelog narratives were aligned to the current 2.4.0 model.
  • super-dev-core, OpenClaw current support, and old skill aliases were removed from the active product surface.

v2.3.8

Patch
  • -SEEAI competition acceptance became stricter: competition delivery now requires more complete and credible runtime and demo evidence before it can count as ready
  • -Evidence content-quality gates added: each section requires ≥8 chars plus template `required` keyword coverage, blocking "ok / done / works" placeholder fills

v2.3.6

Patch
  • -Host onboarding and update flows became more self-healing, with legacy residue, duplicate injections, and stale skills easier to repair automatically
  • -Fixed install selector R-then-Enter failure: skip integrate for hosts without project files

v2.3.5

Patch
  • -SEEAI competition mode became a first-class fast path: shorter and faster than the standard project path, but still grounded in research, document approval, and real runtime proof
  • -SEEAI solution doc replaces three-doc set: single solution.md instead of PRD + Architecture + UIUX for competition scenarios

v2.3.4

Patch
  • -Thanks to staruhub for PR #10, which landed the core orchestration capabilities in this release
  • -Added the Plan-Execute engine with structured plans, topological wave sorting, step-level verification gates, failure budgets, and persisted plan state

v2.3.3

Patch
  • -Claude Code unified to a single super-dev skill entry, eliminating duplicate legacy skill aliases that caused pipeline non-compliance
  • -super-dev update now auto-migrates all host configs after upgrade; super-dev (no args) auto-detects and migrates outdated projects

v2.3.2

Patch
  • -Host product scope was hardened again: 20 unified integration hosts plus 1 manual OpenClaw plugin host, with install scripts, README, website docs, capability audit, and host matrices finally sharing one accurate model
  • -Claude Code and Codex moved further toward their official host contracts: CLAUDE.md + skills + optional plugin enhancement, and AGENTS.md + skills + repo plugin enhancement

v2.3.1

Patch
  • -Codex deep adaptation upgraded to the official AGENTS.md + Skills + repo plugin dual-layer model, with App/Desktop and CLI converging into one Super Dev flow
  • -Claude Code deep adaptation now follows a CLAUDE.md + .claude/skills + ~/.claude/skills + optional plugin enhancement model with unified onboarding, doctor, and runtime semantics

v2.3.0

Major
  • -Governance upgrade: Advisory → Enforcement mode with SKILL.md frontmatter hooks
  • -Three-layer governance: CLAUDE.md persistent rules + Hooks runtime enforcement + CLI on-demand checks
  • -Agent Teams support: parallel expert collaboration across pipeline phases
  • -3-stage error recovery: cheap recovery → context rebuild → surface error
  • -CLAUDE.md include directive: auto-load tech stack knowledge from knowledge/ files

v2.2.0

Major
  • -System-level upgrade: host onboarding, workflow recovery, UI system, quality gates, delivery closure — all rebuilt
  • -Workflow recovery chain: new super-dev resume, bare super-dev auto-enters recovery route, SESSION_BRIEF.md as unified recovery source
  • -Preview confirmation upgraded to first-class gate; next/continue/resume/start/doctor/detect/Web API unified to workflow_state semantics
  • -Host onboarding: 20 unified hosts + 1 OpenClaw manual plugin host, doctor/detect/start with decision cards
  • -Deep host discovery: PATH + common install paths + Windows App Paths + Windows shim + custom SUPER_DEV_HOST_PATH_<HOST>

v2.1.6

Patch
  • -Aligned Python package, CLI, README, website, and host skill surfaces to version 2.1.6
  • -Fixed self-healing refresh for stale host integration files during onboarding

v2.1.5

Patch
  • -Fixed Windows uv tool upgrade file lock issue (auto-retry + clear guidance)
  • -Added super-dev clean command (clean historical artifacts with --keep/--all/--dry-run)

v2.1.3

Patch
  • -OpenClaw plugin upgraded to 13 Tools (added deploy/analyze/doctor)
  • -SKILL.md fully rewritten to 481 lines covering 9-stage execution instructions, 2 gate interaction templates, 3 rework protocols, and resume intelligence

v2.1.2

Major
  • -Added OpenClaw native plugin (@super-dev/openclaw-plugin) with 13 registered tools via Plugin SDK
  • -Expanded the host matrix to 20 unified integration hosts plus the separate OpenClaw plugin host
  • -Fixed 50+ bugs: delete_spec wrong directory, Prisma nullable inverted, TypeORM syntax error, CLI console crash, etc.
  • -Added missing backend/src/app.js entry and frontend/src/App.tsx root component
  • -Added 3 missing backend routes (core/experience/operation)

v2.1.1

Major
  • -Added direct command mode: super-dev "requirement" launches the full pipeline in one line
  • -Added Expert Agent system that auto-activates domain specialists based on project context
  • -Added stage jump command super-dev run <stage_number> to skip completed stages and advance directly
  • -Quality gate now includes A11y accessibility checks (WCAG 2.1 AA) to ensure deliverable accessibility
  • -Upgraded pipeline engine with improved inter-stage state transfer and recovery

v2.1.1

Patch
  • -Focused the public host matrix on 16 primary hosts instead of advertising unstable lab adapters
  • -Added explicit Bugfix Mode through super-dev fix so defect work follows a lightweight patch path

v2.1.0

Major
  • -10 Expert Agent system: PM/ARCHITECT/UI/UX/SECURITY/CODE/DBA/QA/DEVOPS/RCA with full profiles, thinking frameworks, and quality standards
  • -20 host adapters: added Copilot CLI/Kilo Code/Cline/Roo Code, fixed Cursor .mdc/Windsurf rules/Kiro steering formats
  • -119 color schemes (with dark mode auto-generation) + 39 component libraries + 17 font presets
  • -Added direct command mode super-dev "requirement" and stage jump super-dev run 1-9

v2.0.12

Patch
  • -Fixed version alignment and release pipeline
  • -Improved PyPI package build and metadata

v2.0.11

Patch
  • -UI Intelligence major expansion: color schemes from 7 to 84 + 35 aesthetic directions
  • -Component library recommendations expanded from 12 to 39 covering React/Vue/Angular/Svelte/Mini Programs/RN/Flutter

v2.0.10

Patch
  • -Added explicit bootstrap artifacts (.super-dev/WORKFLOW.md and output/*-bootstrap.md)
  • -Added integrate validate for host runtime acceptance with persisted status

v2.0.9

Patch
  • -Added Impact Analysis for change scope assessment
  • -Added Feature Checklist for PRD coverage audit

v2.0.8

Patch
  • -Fixed host detection issues in edge environments
  • -Improved pipeline-contract output format

v2.0.7

Patch
  • -Fixed skill install path resolution, slash command registration, and host detection edge cases
  • -Improved CI/CD YAML generation format

v2.0.6

Patch
  • -Fixed Windsurf rules path and Kiro steering frontmatter format
  • -Improved host rule file content consistency

v2.0.5

Patch
  • -Fixed Cursor .mdc frontmatter compatibility
  • -Improved Codex CLI Flow Contract injection

v2.0.4

Patch
  • -Improved host adapter detection logic and error messages
  • -Fixed slash command paths for multiple hosts

v2.0.3

Patch
  • -Host adapter expansion: added Kilo Code, Cline, Roo Code
  • -Skill-only mode: hosts can install skill without slash command

v2.0.2

Patch
  • -Host-scoped workflow isolation: independent protocol surfaces per host
  • -Fixed file conflicts during concurrent multi-host onboarding

v2.0.1

Patch
  • -Finalized workflow documentation and task-closure hardening
  • -Fixed spec task status persistence

v2.0.0

Major
  • -Rebuilt the 12-stage pipeline engine with run --resume support
  • -Added enterprise policy preset
  • -Added full delivery archive packaging
  • -Added host profile and compatibility scoring
  • -Added three-dimensional red-team review

v1.0.1

Patch
  • -Added complete workflow tutorial (WORKFLOW_GUIDE.md)
  • -Integrated design intelligence engine (colors, fonts, charts)

v1.0.0

Major
  • -Initial release
  • -Base pipeline framework (research / documents / spec / implement)
  • -Spec-Driven Development module
  • -Support for Claude Code, Cursor, and Windsurf
  • -Official PyPI release