super-dev:基于 Python 的 AI 开发编排工具项目

https://github.com/shangyankeji/super-dev的镜像

Branch3Tags22
FilesLast commitLast update
6 months ago
6 months ago
6 months ago
7 months ago
7 months ago
6 months ago
6 months ago
6 months ago
7 months ago
6 months ago
6 months ago
7 months ago
6 months ago
6 months ago
7 months ago
6 months ago
10 months ago
7 months ago
7 months ago
6 months ago
7 months ago
6 months ago
8 months ago
6 months ago
6 months ago
6 months ago
9 months ago
6 months ago
7 months ago
10 months ago
8 months ago
8 months ago
6 months ago
10 months ago
6 months ago
7 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
7 months ago
10 months ago
6 months ago
6 months ago
7 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago
6 months ago

Super Dev

Super Dev - AI PIPELINE ORCHESTRATOR

Host Coaching System for Commercial-Grade Delivery

License: MIT Python Type Checks Lint

简体中文 | English


Version

Current version: 2.4.0

Install

The homepage installation story is now intentionally uv-first:

uv tool install super-dev

Then run:

super-dev

Source installs and rollback/version-pinning flows remain in docs/INSTALL_OPTIONS.md.

Cross-platform note:

  • Windows, macOS, and Linux all use the same entry: install the package, then run super-dev.
  • The repository install.sh is a macOS/Linux convenience script, not the only Windows path.

What's New in 2.4.0

Public Surface, Finally Host-First

  • The terminal is now intentionally small:
    • super-dev
    • super-dev update
    • super-dev uninstall
  • Real delivery work happens inside the host through:
    • /super-dev
    • /super-dev-seeai
    • continue current flow
    • what is the next step now
  • super-dev-core has been removed from current code, templates, skills, and active contracts. The product name is consistently super-dev.

Existing-Project Flow Is Now First-Class

  • Existing-project work now follows a formal chain instead of pretending every job is greenfield:
    • baseline -> baseline confirmation -> delta research -> docs -> docs_confirm -> spec -> frontend -> preview_confirm -> backend -> quality -> delivery
  • resume is a first-class scenario, not an exception.
  • workflow_context, baseline_governance, and resume_gate now drive detect / doctor / validate / proof-pack / release-readiness consistently.

26-Host Unified Matrix

  • The current matrix is now unified as:
    • 12 CLI hosts
    • 9 IDE hosts
    • 5 desktop assistants
  • Droid CLI, Kimi Code, Qwen Code, Trae SOLO, and Trae SOLOCN are first-class hosts in the current matrix.
  • OpenClaw is no longer part of the current support surface.
  • Host adaptation now includes:
    • project-first onboarding
    • optional user/global surfaces
    • official alignment
    • host start playbooks
    • resume guidance
    • repair playbooks
    • post-onboard self-checks
    • standard vs. SEEAI readiness

Project-First Onboarding

  • Super Dev no longer defaults to writing rules into system-level AGENTS.md.
  • The default is now project-first:
    • write project surfaces first
    • only add user/global surfaces with explicit opt-in
  • This keeps onboarding safer, cleaner, and closer to real team workflows.

UI Design Upgrade: Less AI Slop, More Intentional Design

  • The UI contract now freezes:
    • art-direction candidates
    • a visual philosophy / direction manifest
    • anti-AI-slop guardrails
    • a five-dimension critique rubric
    • tweak categories
  • Frontend blueprints and UI review now consume those signals directly instead of treating design as a generic template problem.

Release Gates Are Green Again

  • pytest -q: 2576 passed, 2 skipped
  • scripts/preflight.sh --allow-dirty: passing
  • Core release gates, knowledge gates, delivery smoke, host compatibility, build, and twine checks are green again

For the full breakdown, see docs/releases/2.4.0.md.


Demo Video


What is Super Dev

Super Dev is not another layer of low-level commands or a scaffolding shell. It is a host coaching system that trains the model capabilities inside your host into a stable, transparent, auditable delivery flow for real commercial software work.

Division of responsibility:

  • The host handles model inference, web research, code generation, terminal execution, and file modifications.
  • Super Dev handles blueprints, workflow governance, design constraints, quality gates, audit artifacts, and delivery standards.

In practice:

  • the host is the executor
  • Super Dev is the product coach, architecture coach, design director, and QA gatekeeper

Problems it solves:

  • Converts requirements into production artifacts: PRD, architecture, UI/UX spec, task plans, and delivery manifests.
  • Organizes development into a standardized pipeline: traceable, resumable, auditable, and reviewable.
  • Enforces quality at every stage: policy governance, red-team review, quality gates, and release rehearsals.
  • Unifies collaboration across 12 CLI hosts, 9 IDE hosts, and 5 desktop assistants under one project-first delivery standard.
  • Trains the host into a more reliable commercial-delivery team instead of leaving users to manually improvise process, design, and quality.

5-Minute Start

Regular users should remember 1 install command and 3 terminal commands:

uv tool install super-dev
# Enter the host onboarding flow
super-dev

# Upgrade to the latest version and migrate onboarded hosts
super-dev update

# Remove injected Super Dev surfaces from onboarded hosts
super-dev uninstall

5-minute path

  1. Run super-dev inside the project directory.
  2. Let the installer detect hosts and write project-level surfaces first; only add user/global surfaces when you explicitly want them.
  3. Open output/maintenance/host-onboard-smoke-*.md.
  4. Read the standard-flow first prompt or competition-flow first prompt.
  5. Check post-onboard self-check, Framework focus, and official workflow checks.
  6. Go back to the host and trigger the first prompt. The host should enter research -> three core docs -> wait for confirmation.

If those six steps hold, you did not just “install a tool.” You turned the current host into a coached delivery system for this repository.

The terminal only does 3 things

  • onboarding
  • upgrade
  • uninstall / cleanup preview

After onboarding, regular use moves back into the host. Regular users should only need to remember:

/super-dev your requirement
super-dev: your requirement
/super-dev-seeai your competition brief

Do not try to memorize every host-specific trigger. Open the generated host-onboard-smoke guide and copy the standard-flow first prompt or competition-flow first prompt.

Correct mental model:

  • The terminal is only for onboarding, upgrading, and uninstalling.
  • The host is where baseline, research, the three core docs, approval gates, spec, implementation, quality gates, and delivery happen.
  • Auto-judgement is allowed during onboarding and upgrade, not during normal development flow.
  • Integrated and runtime verified are different states. Files existing does not prove the host actually follows the governed flow.
  • Existing-project work must not jump directly into docs or coding. evolve / variant / patch always start with a baseline scan.
  • Resume is the default scenario. Closing the host, shutting the laptop, or returning the next day should continue from .super-dev/ and output/ artifacts.

Work modes:

  • new: greenfield 0-to-1 delivery
  • evolve: additive work on the current product
  • variant: 1-N+1 derivative based on the current product
  • patch: bugfix / remediation on the current product
  • resume: continue the interrupted workflow

Recommended first run:

  1. Run super-dev in the terminal.
  2. Let the installer detect hosts and write the required project-level surfaces by default; add --with-user-surfaces only when you explicitly want cross-project user/global protocol surfaces.
  3. The installer is project-first: it writes project surfaces before optional user/global surfaces.
  4. Open output/maintenance/host-onboard-smoke-*.md and validate the host using the generated smoke guide first, including any Framework focus when the project already has a frozen framework playbook.
  5. Open the host and use /super-dev or super-dev:.
  6. For existing projects, let the host finish baseline first, then generate delta research, PRD, Architecture, and UI/UX.
  7. Only move to Spec and implementation after the docs are confirmed.

Main rule: once installation is done, the terminal should mostly step aside. Real project work moves back into the host. Copy the standard-flow or competition-flow first prompt first, then follow the smoke guide.

SEEAI competition fast mode:

  • Entry: /super-dev-seeai or super-dev-seeai:
  • It still keeps research -> docs -> docs confirm -> spec
  • After Spec it goes straight into an integrated full-stack sprint, without a separate preview confirmation gate
  • Best for 30-minute showcase builds such as a polished landing page, mini-game, or focused demo tool

Everything else (detect / doctor / review / quality / release / spec / task) remains available as maintenance surface only. Regular users should only need:

super-dev
super-dev update
super-dev uninstall

Maintainer additions:

  • Post-onboard smoke guides are written to output/maintenance/host-onboard-smoke-*.md
  • If the project already has a frozen framework playbook, the smoke guide also exposes Framework focus, validation surfaces, and delivery evidence.
  • Uninstall previews and cleanup audits are written to output/maintenance/host-cleanup-*.json/.md
  • Use super-dev uninstall --dry-run to preview removals before deleting anything
  • After upgrade, reopen the terminal and run super-dev --version; if the host no longer enters research -> three core docs -> wait for confirmation, then return to the terminal and run doctor
  • If the first UI pass still looks flat, empty, or template-like, do not defer polish to the end. A screenshot-level UI gate now keeps blocking that state across UI review, quality gate, proof-pack, and release readiness.
  • docs/PUBLISHING.md and docs/RELEASE_RUNBOOK.md are maintainer-only release docs, not part of the normal end-user path.

Core Features

1. 11-Expert Agent Architecture

Super Dev currently ships with eleven domain-expert agents. Each expert is injected into prompts at the corresponding pipeline stage so the host stays constrained to professional-grade output:

Expert Role Injection Stages
PRODUCT Product Lead research, prd, quality, delivery
PM Product Manager research, prd
ARCHITECT System Architect architecture
UI Interface Designer uiux, frontend
UX Interaction Designer uiux, frontend
SECURITY Security Engineer architecture, backend, quality
CODE Software Engineer frontend, backend
DBA Database Architect architecture, backend
QA Quality Assurance quality
DEVOPS DevOps Engineer delivery
RCA Root Cause Analyst quality, delivery

Each expert carries: objective definition, background story, thinking framework, and quality criteria. The generated AI prompts ensure every stage meets domain-specific professional baselines.

2. UI Design Intelligence System

A built-in design intelligence engine that directly constrains visual quality during frontend implementation:

  • 119 color palettes: 84 product palettes + 35 aesthetic palettes, all with automatic dark mode generation.
  • 39 component libraries: covering 11 frontend stacks (React 15 / Vue 9 / Angular 4 / Svelte 2 / others).
  • 17 typography presets: based on Google Fonts, categorized by product tone and personality.
  • Complete design token system: color scales, shadows, motion, typography, and spacing.
  • 12-item pre-delivery checklist: A11y, responsive design, dark mode, loading states, empty states, error states, and more.
  • 10 industry customizations: education, healthcare, e-commerce, fintech, SaaS, social, content, enterprise, utilities, and gaming.

The UI system is no longer only advisory. It is frozen into actual artifacts:

  • output/*-uiux.md
  • output/*-ui-contract.json
  • output/frontend/design-tokens.css
  • output/*-ui-contract-alignment.md
  • output/*-ui-contract-alignment.json

Host prompts, implementation constraints, UI review, frontend runtime, quality gate, proof-pack, and release readiness all consume that same UI contract.

Key governance evidence:

  • UI review and UI contract alignment
  • Frontend runtime evidence
  • Proof-pack and release readiness summaries

3. Pipeline Orchestration Engine

  • 9-stage standard pipeline: research -> prd -> architecture -> uiux -> spec -> frontend -> backend -> quality -> delivery.
  • Checkpoint and resume: interrupted pipelines resume from the last completed stage without losing progress.
  • Stage timeout protection: each stage has a timeout mechanism to prevent indefinite stalling.
  • Confirmation gates: mandatory user confirmation after core documents and after frontend preview.
  • Stage recovery: natural-language recovery inside the host is the default path; explicit stage-jump controls remain maintenance-only.
  • UI revision loop: when the frontend needs another pass, start the revision loop inside the host and update UIUX before redoing the frontend.
  • Dual-mode delivery: works for both greenfield (0-1) and iterative (1-N+1) projects.
  • Continuation routing: internal recovery and status commands share the same workflow state and action card semantics.
  • Session recovery card: .super-dev/SESSION_BRIEF.md and .super-dev/workflow-state.json persist the current action, host first sentence, machine action, and continuity rules.
  • Recent operational timeline: workflow snapshots, semantic workflow events, and hook events are merged into one timeline that now surfaces in SESSION_BRIEF, Workflow Harness, proof-pack, and release readiness.
  • Rework-first state handling: docs confirmation, preview confirmation, UI redesign, architecture rework, and quality remediation all stay inside one governed state machine.

4. Document Generation Engine

Super Dev generates an initial document framework for each stage. The host LLM then enriches it with user requirements, web research, and expert knowledge:

Document Content
PRD User personas, feature matrix, acceptance criteria, competitive benchmarking, business rules
Architecture System architecture, data models, API contracts, security strategy, deployment plan
UIUX Design tokens, page skeletons, component inventory, interaction states, responsive strategy

The host expands documents based on actual project needs. Final document scope depends on project complexity. Supports 10 industry-specific customizations: education, healthcare, e-commerce, fintech, SaaS, social, content, enterprise, utilities, and gaming.

5. Quality Gate System

  • A11y accessibility checks.
  • Performance budget enforcement.
  • Red-team review (security / performance / architecture).
  • Fix command suggestions (detected issues produce actionable repair instructions).
  • Policy governance (default / balanced / enterprise presets).
  • Spec quality scoring and release-readiness panel.
  • UI contract execution checks (ui-contract.json, design-tokens.css, frontend runtime, and UI alignment evidence must stay consistent).

6. Host Onboarding Governance

  • A unified onboarding matrix grouped into CLI hosts, IDE hosts, and desktop assistants.
  • Droid CLI joins the official host matrix via AGENTS.md + .factory/rules + .factory/skills, with .factory/commands kept as a compatibility enhancement.
  • Auto-generates host rule files, slash command mappings, and Skill directories.
  • Host capability boundary modeling: Certified / Compatible / Experimental three-tier certification.
  • detect / onboard / doctor / setup / install / start form a closed onboarding loop.
  • --dry-run preview mode and --stable-only stable-only mode.
  • doctor, detect, and start now emit decision cards: recommended host, recommendation reason, first action, folded candidate list, and path-override hints.
  • Decision cards, the installer, and runtime reports now surface the host's standard-flow first prompt, competition-flow first prompt, post-onboard self-check, official workflow checks, and repair playbook.
  • Host readiness is no longer just "files exist"; it explicitly distinguishes "ready for standard flow" vs. "ready for SEEAI competition flow".
  • Supports Windows registry hits, shim directories, common install paths, and SUPER_DEV_HOST_PATH_<HOST> explicit path overrides.
  • If you explicitly choose a host, Super Dev centers guidance around that host instead of letting auto-detection override your intent.

7. Internal Analysis and Scope Governance

The runtime still keeps internal analysis capabilities such as scope coverage, blast-radius estimation, and regression planning. They remain part of the governance core, but they are no longer taught as standalone public commands for ordinary users.

8. Auditable Delivery

  • pipeline-metrics: telemetry and metrics report.
  • pipeline-contract: stage-level contract evidence.
  • resume-audit: resume execution audit trail.
  • delivery manifest/report/archive: delivery package.
  • proof-pack: delivery evidence bundle with executive summary.
  • release readiness, Spec Quality, and Scope Coverage: unified release scoring panel.
  • UI Contract Alignment is part of proof-pack and release readiness, not just an internal UI review detail.
  • Governance snapshots, frontend runtime, validation reports, and knowledge tracking now participate in delivery closure.

9. Knowledge Base

  • Project-level knowledge/ directory for domain knowledge files.
  • Knowledge bundle caching at output/knowledge-cache/*-knowledge-bundle.json.
  • Matched local standards, scenario packs, and checklists are treated as hard constraints.
  • Hosts must read relevant knowledge files before drafting PRD, architecture, and UIUX documents.
  • Knowledge hits are inherited into Spec and implementation stages.

10. Policy Governance (Policy DSL)

Parameterized workflow governance through a Policy DSL:

  • default: standard preset, suitable for individuals and small teams.
  • balanced: balanced preset, suitable for medium-sized teams.
  • enterprise: enterprise preset, higher quality thresholds, host profiling requirements, and configurable required hosts per project.

Governance control dimensions:

  • Mandatory red-team / quality gate toggle.
  • Minimum quality threshold enforcement.
  • CI/CD platform whitelist.
  • Required hosts and ready+score hard validation (enabled per project).
  • Automatic host profiling and scoring.
  • host-compatibility report with history tracking.

Configurable via super-dev.yaml policy section.


How It Works

  1. User runs super-dev in the project directory.
  2. The onboarding wizard connects Super Dev to the target host.
  3. User triggers Super Dev using the host's supported entry, such as /super-dev requirement, super-dev: requirement, Codex App/Desktop selecting super-dev from the / list, or Codex CLI typing $super-dev.
  4. The host enters the Super Dev pipeline; 11 expert agents are injected by stage.
  5. The host handles web research, inference, coding, execution, and file modifications.
  6. Super Dev handles workflow, documents, gates, audit, and delivery standards.

Standard flow for new work: research -> docs -> docs_confirm -> spec -> frontend -> preview_confirm -> backend -> quality -> delivery

Existing-project flow: baseline -> baseline confirmation -> delta docs -> docs_confirm -> spec -> frontend -> preview_confirm -> backend -> quality -> delivery

New features follow the full pipeline. Bug fixes follow a lightweight patch path (symptoms, reproduction, blast radius, regression risk) without skipping documentation. Analysis stages automatically exclude .venv, site-packages, node_modules, and other non-source directories.

How Hosts Understand Super Dev

  • Super Dev is a local Python CLI tool plus host-side rule files / Skills / slash mappings.
  • The host handles inference, research, coding, and execution. Super Dev handles pipeline flow, gates, and audit.
  • When the user uses the host-supported Super Dev entry (/super-dev, super-dev:, Codex App/Desktop /-list skill entry, or Codex CLI $super-dev), the host switches to pipeline mode.
  • If a knowledge/ directory exists, the host reads relevant knowledge files before drafting documents.
  • If output/knowledge-cache/*-knowledge-bundle.json exists, its knowledge hits are inherited into all later stages.

Installation

uv tool install super-dev

Upgrade:

super-dev update

2. Pin a specific version

uv tool install super-dev==2.4.0

Upgrade:

uv tool upgrade super-dev
super-dev update

After installation, run super-dev to launch the interactive host onboarding wizard (Up/Down to navigate, Space to select, Enter to install, A for all, C for CLI only, I for IDE only, R to reset). The terminal prints the exact trigger command for each selected host.

To explicitly initialize the project contract before onboarding:

super-dev bootstrap --name my-project --platform web --frontend next --backend node

This generates .super-dev/WORKFLOW.md and output/*-bootstrap.md to lock down initialization spec, trigger method, and stage sequence.

3. Install from GitHub tag

uv tool install --from git+https://github.com/shangyankeji/super-dev.git@v2.4.0 super-dev

4. Source install for development

git clone https://github.com/shangyankeji/super-dev.git
cd super-dev
uv sync
uv run super-dev --version

Dependency Notes

uv automatically installs Super Dev's own Python dependencies (rich, pyyaml, ddgs, requests, beautifulsoup4, fastapi, uvicorn, etc.).

It does not install:

  • Host applications (Claude Code, Codex CLI, Gemini CLI, Cursor, Trae, Windsurf, and desktop assistants)
  • System runtimes (Node.js, npm, pnpm, Docker, database services)
  • Host authentication state, browsing permissions, or API keys
  • Project-specific frontend/backend runtime dependencies

In short: uv installs Super Dev's Python dependencies. It does not install host tools or system environments on your behalf.


Architecture Overview

System Flow Architecture

Shows the relationship between users, host-side tools, the Super Dev orchestration engine, and final artifacts.

System Overview

Pipeline Stage Flow

Details the internal execution flow after each host-side trigger.

Pipeline Stages

Core Module Topology

Shows the responsibility boundaries and call relationships of core source directories under super_dev.

Module Topology


Unified Host Matrix

Super Dev officially documents one unified onboarding matrix:

  • Certified: fully aligned integration model; recommended for production use.
  • Compatible: complete integration path; awaiting extended real-world validation.
  • Experimental: functional integration; needs broader production testing.
  • Project-first injection is the default: project surfaces are written first, then optional user/global surfaces.
  • Codex and Codex CLI are distinct hosts, and Claude and Claude Code are distinct hosts.

Unified CLI Hosts (12)

Host Trigger Terminal Entry
Claude Code /super-dev your requirement super-dev
Codex CLI $super-dev; fallback super-dev: your requirement super-dev
OpenCode /super-dev your requirement super-dev
Droid CLI /super-dev your requirement; competition mode /super-dev-seeai; fallback super-dev: your requirement super-dev
Gemini CLI /super-dev your requirement super-dev
Kiro CLI /super-dev your requirement super-dev
Cursor CLI super-dev: your requirement super-dev
Copilot CLI super-dev: your requirement super-dev
Qoder CLI /super-dev your requirement super-dev
CodeBuddy CLI /super-dev your requirement super-dev
Kimi Code super-dev: your requirement; explicit entry /skill:super-dev your requirement super-dev
Qwen Code /super-dev your requirement super-dev

IDE Hosts (9)

Host Trigger Terminal Entry
Antigravity /super-dev your requirement super-dev
Cursor /super-dev your requirement super-dev
Windsurf /super-dev your requirement super-dev
Kiro /super-dev your requirement super-dev
Trae IDE super-dev: your requirement super-dev
TraeCN super-dev: your requirement super-dev
CodeBuddy /super-dev your requirement super-dev
CodeBuddyCN /super-dev your requirement super-dev
Qoder /super-dev your requirement super-dev

Desktop Assistants (5)

Host Trigger Terminal Entry
Claude super-dev: your requirement super-dev
Codex App/Desktop: select super-dev from the / list super-dev
WorkBuddy super-dev: your requirement super-dev
Trae SOLO /super-dev your requirement super-dev
Trae SOLOCN super-dev: your requirement super-dev

Public Terminal Entry

super-dev
super-dev update

Run super-dev to open the installer and choose the target host there. onboard --host, doctor, and other CLI commands remain available for maintenance, but they are no longer the normal end-user path.

Per-Host Usage Details

CLI Hosts

Claude Code

super-dev

Trigger location: launch Claude Code in the project directory, then trigger within the same session. Trigger command: /super-dev your requirement Fallback: super-dev: your requirement Restart required after onboarding: No.

Notes:

  1. Recommended as the primary CLI host.
  2. If a maintainer needs to verify the installed surfaces, run super-dev doctor --host claude-code to confirm project-root CLAUDE.md, project .claude/CLAUDE.md, optional .claude/settings*.json, and project/user skills plus agents are active together.
  3. Claude Code is now aligned to the official CLAUDE.md + settings + project/user skills + subagents model; .claude/commands/ remains compatibility-only rather than the primary contract.
  4. Super Dev can also install .claude-plugin/marketplace.json plus plugins/super-dev-claude/.claude-plugin/plugin.json as an optional repo-local Claude plugin enhancement.

Codex

super-dev

Trigger location: after onboarding, restart codex, then trigger in the new session. Trigger command: Codex App/Desktop: choose super-dev from the / list Codex CLI: $super-dev Fallback: super-dev: your requirement Restart required after onboarding: Yes.

Notes:

  1. In Codex app/desktop, prefer selecting super-dev directly from the / list; that is the enabled Skill entry, not a project-level custom slash command.
  2. In Codex CLI, prefer explicit $super-dev.
  3. If you are already continuing inside natural-language context, you can still use super-dev: your requirement as the AGENTS-driven fallback.
  4. The default base surfaces are project AGENTS.md plus project .agents/skills/super-dev/SKILL.md; the official user-level Skill at ~/.agents/skills/super-dev/SKILL.md is still installed, while CODEX_HOME/AGENTS.md (default ~/.codex/AGENTS.md) is now only written when --with-user-surfaces is explicitly requested.
  5. Super Dev also installs an optional repo-local Codex plugin enhancement at .agents/plugins/marketplace.json + plugins/super-dev-codex/.codex-plugin/plugin.json so Codex App/Desktop can expose a richer local plugin surface alongside AGENTS + Skills.
  6. Older installations are migrated forward to the unified super-dev naming during upgrade.
  7. If a previous session did not load the new surfaces, restart codex and try again.

Gemini CLI

super-dev

Trigger location: launch Gemini CLI in the project directory. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. The official primary model is now GEMINI.md + settings + custom commands; inside the project, check GEMINI.md, optional .gemini/settings.json, and .gemini/commands/*.toml first.
  2. /super-dev is the injected Super Dev custom command, not a Gemini-native built-in command; if the command list has not refreshed, reopen the Gemini CLI session first.
  3. ~/.gemini/skills/ is kept only as a compatibility enhancement, not as the default primary protocol surface.
  4. Complete the full pipeline within a single session: research -> three documents -> user confirmation -> Spec -> frontend verification -> backend / delivery.

Cursor CLI

super-dev

Trigger location: launch Cursor CLI in the project directory. Trigger command: super-dev: your requirement Restart required after onboarding: No.

Notes:

  1. Suitable for continuous research, documentation, and coding within the terminal.
  2. The official project context model is project-root AGENTS.md + .cursor/rules/; root CLAUDE.md remains compatibility context only.
  3. If project context or rules have not refreshed, reopen the Cursor CLI session; when resuming an existing workflow, prefer Cursor CLI's native session continuity instead of starting a new thread.

Kiro CLI

super-dev

Trigger location: launch Kiro CLI in the project directory. Trigger command: /super-dev your requirement Restart required after onboarding: Yes.

Notes:

  1. Kiro CLI now prefers the currently exposed /super-dev host entry; if the session only accepts natural language, fall back to super-dev: your requirement.
  2. The official integration surfaces are AGENTS.md + .kiro/steering/super-dev.md + .kiro/skills/super-dev/SKILL.md; global steering and skills stay as optional enhancement surfaces.
  3. Relaunch Kiro CLI after onboarding so steering context and skills load in the new session.

OpenCode

super-dev

Trigger location: launch OpenCode in the project directory. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. Uses CLI slash mode.
  2. If you use a global command directory, keep the project-level onboarding files as well.

Qoder CLI

super-dev

Trigger location: launch Qoder CLI in the project directory. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. Suitable for command-line pipeline development.
  2. If slash is not active, confirm that AGENTS.md and .qoder/commands/super-dev.md exist and that the .qoder/rules/ directory has been created.
  3. The official surfaces now use AGENTS.md + .qoder/rules/super-dev.md + .qoder/commands/super-dev.md + .qoder/skills/ / ~/.qoder/AGENTS.md + ~/.qoder/commands/ + ~/.qoder/skills/; .qoder/agents/ stays as an enhancement layer only.

CodeBuddy CLI

super-dev

Trigger location: launch CodeBuddy CLI in the project directory. Trigger command: /super-dev your requirement Competition mode: /super-dev-seeai your competition brief Restart required after onboarding: No.

Notes:

  1. Type the command directly in the current CLI session.
  2. The official primary surfaces are CODEBUDDY.md + .codebuddy/rules/ + .codebuddy/commands/ + .codebuddy/skills/ + .codebuddy/agents/, plus ~/.codebuddy/CODEBUDDY.md.
  3. If the session was opened before onboarding, reload project rules before triggering.
  4. For hackathon-style work, prefer /super-dev-seeai so the host stays on the 30-minute fast path.

IDE Hosts

Antigravity

super-dev

Trigger location: open the Agent Chat / Prompt panel in Antigravity with the project workspace active. Trigger command: /super-dev your requirement Restart required after onboarding: Yes.

Notes:

  1. Uses the GEMINI.md + custom commands integration model, with .agent/workflows kept as the recommended enhancement layer.
  2. Onboarding writes project-level GEMINI.md, .gemini/commands/super-dev.toml, and .agent/workflows/super-dev.md.
  3. By default it only writes project-level GEMINI.md and project command surfaces; user-level ~/.gemini/GEMINI.md and ~/.gemini/commands/ are only written when --with-user-surfaces is explicitly requested, while ~/.gemini/skills/ remains a compatibility enhancement only.
  4. After onboarding, reopen Antigravity or start a new Agent Chat before triggering.

Cursor IDE

super-dev

Trigger location: open Agent Chat in Cursor with the target project as the active workspace. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. Complete the full pipeline within a single Agent Chat session.
  2. If project rules did not load, reopen the workspace or start a new chat.

Windsurf

super-dev

Trigger location: open Agent Chat or the Workflow entry in Windsurf within the project context. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. Uses the AGENTS.md + rules + workflow + skills integration model.
  2. Best suited for completing research, documents, Spec, and coding within a single Workflow.
  3. The official docs expose AGENTS.md, .windsurf/workflows/, and .windsurf/skills/; the repo keeps .windsurf/rules/ as the project constraint layer.

Kiro IDE

super-dev

Trigger location: open the Agent Chat / AI panel in Kiro IDE within the project context. Trigger command: /super-dev your requirement Restart required after onboarding: Yes.

Notes:

  1. Uses the currently exposed /super-dev host entry first; if the current session only accepts natural language, fall back to super-dev: your requirement.
  2. Onboarding writes project-level .kiro/steering/super-dev.md and .kiro/skills/super-dev/SKILL.md, plus global ~/.kiro/steering/super-dev.md and ~/.kiro/skills/super-dev/SKILL.md; legacy ~/.kiro/steering/AGENTS.md remains as a compatibility surface.
  3. If steering or skills are not loaded, reopen the project window or start a new Agent Chat.

Qoder IDE

super-dev

Trigger location: open Agent Chat in Qoder IDE within the current project. Trigger command: /super-dev your requirement Restart required after onboarding: No.

Notes:

  1. Uses the official AGENTS.md + commands + rules + skills project model; type /super-dev your requirement directly in Agent Chat.
  2. If the new command does not appear, confirm AGENTS.md, .qoder/commands/super-dev.md, and .qoder/rules/super-dev.md exist, then reopen the project or start a new Agent Chat.
  3. The official surfaces now use AGENTS.md + .qoder/rules/super-dev.md + .qoder/commands/super-dev.md + .qoder/skills/ / ~/.qoder/AGENTS.md + ~/.qoder/commands/ + ~/.qoder/skills/; .qoder/agents/ stays as an enhancement layer only.

Trae IDE

super-dev

Trigger location: open Agent Chat in Trae IDE within the current project context. Trigger command: super-dev: your requirement Restart required after onboarding: No.

Notes:

  1. Uses super-dev: your requirement as the primary trigger.
  2. By default onboarding writes project-level .trae/project_rules.md and .trae/rules.md; user-level ~/.trae/user_rules.md and ~/.trae/rules.md are only written with explicit --with-user-surfaces; if a compatible skill directory is detected, it also installs ~/.trae/skills/super-dev/SKILL.md.
  3. After onboarding, reopen Trae or start a new Agent Chat to activate rules; if the compatible Skill was installed, it activates as well.
  4. Proceed using output/* and .super-dev/changes/*/tasks.md.

CodeBuddy

super-dev

Trigger location: open Agent Chat in CodeBuddy within the project context. Trigger command: /super-dev your requirement Competition mode: /super-dev-seeai your competition brief Restart required after onboarding: No.

Notes:

  1. Use within a project-level Agent Chat; do not leave the project context.
  2. Let the host complete research before proceeding to documents and coding.
  3. Uses CODEBUDDY.md + .codebuddy/rules/super-dev/RULE.mdc + .codebuddy/commands/ + .codebuddy/agents/ + .codebuddy/skills/ integration surfaces.
  4. For competition sprints, keep the same Agent Chat alive and avoid unnecessary sub-session switching.

Copilot (VS Code) / Roo Code / Kilo Code / Cline

These IDE hosts and desktop assistants all use the same pattern: run super-dev in the terminal, choose the target host in the installer, then trigger with super-dev: your requirement inside the host chat or workspace. No restart is required after onboarding unless the host says so.

super-dev

Droid CLI

Droid CLI uses Factory's official host surfaces rather than a separate plugin package.

Integration surfaces:

  • AGENTS.md
  • .factory/rules/super-dev.md
  • .factory/skills/super-dev/SKILL.md
  • .factory/skills/super-dev-seeai/SKILL.md
  • compatibility enhancement: .factory/commands/super-dev.md, .factory/commands/super-dev-seeai.md
  • user-level ~/.factory/AGENTS.md, ~/.factory/commands/, ~/.factory/skills/ (not written by default; only when --with-user-surfaces is explicitly requested)

Installation:

super-dev

Choose Droid CLI in the installer, then go back to the active Droid project session.

Trigger:

/super-dev your requirement

Fallback:

super-dev: your requirement

Competition mode:

/super-dev-seeai your competition brief

Headless resume:

droid exec --session-id <id> "continue with next steps"

Notes:

  1. Keep Droid in the same project session so it reuses the current workflow context instead of restarting.
  2. /super-dev-seeai is the preferred SEEAI competition entry; use super-dev-seeai: only as a fallback.
  3. Droid's primary official model is AGENTS.md + .factory/rules + .factory/skills; .factory/commands/ remains a compatibility enhancement alongside that core model.

Internal / Maintenance Surface

Support, governance, and advanced maintenance commands still exist, but they are no longer the public end-user path.

The public surface stays at:

  • super-dev
  • super-dev update
  • super-dev uninstall
  • /super-dev ...
  • /super-dev-seeai ...
  • continue current flow
  • what is the next step now

Everything else should be treated as maintenance-only:

  • doctor / detect / onboard / skill / integrate for host injection and validation
  • review / quality / release for governance evidence and gate synchronization
  • spec / task for maintainer-side change management

For interrupted work, prefer host-side recovery first:

  • continue current flow
  • baseline confirmed, continue current flow
  • docs confirmed, continue current flow

Documentation

Execution principles:

  • The host is responsible for "writing code."
  • Super Dev is responsible for "making the development process correct, complete, and auditable."

License

MIT

Introduction

https://github.com/shangyankeji/super-dev的镜像

Customize your domain