web-builder
Universal direct-build website and web-app engineering skill
Open source Open in the app JSON README (API)
About
Universal direct-build website and web-app engineering skill
Details
- Kind
- Plugins
- Topic
- No topic detected
- Publisher
- rishav7324
- Origin
- gemini
- Category
- ferramentas
- Version
- 7.1.0
- Last push
- 2026-09-03T13:38:15Z
- Repository state
- ativo
- Language
- Python
- License
- MIT
- Added
- 2026-09-03 22:01:09
- Updated
- 2026-09-03 22:01:09
- Origin id
rishav7324/web-builder
README
# Web Blueprint Engine — Universal Direct Build v7 A portable, production-oriented Agent Skill for turning website and web-app requirements into **real, verified code**. ## What v7 adds - One canonical `skill/SKILL.md` - Agent Skills-compatible layout - Claude Code fallback via `CLAUDE.md` + `.claude/skills/...` - OpenCode discovery via `.opencode/skills/...` and `.agents/skills/...` - Codex-compatible repository guidance via `AGENTS.md` + `.agents/skills/...` - Google Antigravity / Agent Skills-oriented `.agents/skills/...` - Claude-compatible plugin metadata in `.claude-plugin/plugin.json` - Cross-platform Python, shell and PowerShell installers - Non-destructive agent detection - Project-local installation by default - Explicit global installation - Safe `--force` overwrite only when intentionally requested - Mirror verification and synchronization utilities - Multi-agent repository safety rules - Original reference material retained under `references/` - Direct implementation instead of mandatory PRD/TRD/prompt generation ## Repository layout ```text skill/SKILL.md # canonical skill references/ # original V4-derived detailed references .agents/skills/web-blueprint-engine/ # Agent Skills / Codex / Antigravity mirror .claude/skills/web-blueprint-engine/ # Claude Code mirror .opencode/skills/web-blueprint-engine/ # OpenCode mirror .claude-plugin/plugin.json # Claude-compatible plugin metadata AGENTS.md # generic/Codex repository fallback CLAUDE.md # Claude Code repository fallback installer/ # detection, install, sync, verify agent-adapters/ # machine-readable interoperability policy skill-manifest.json # package metadata ``` ## Install for one or many agents ### Automatic project-local detection ```bash python3 installer/install.py --project --targets auto ``` ### Install all project-local adapters ```bash python3 installer/install.py --project --targets all ``` ### Global installation (explicit) ```bash python3 installer/install.py --global --targets all ``` ### Detect the host ```bash python3 installer/install.py --detect ``` ### Windows PowerShell ```powershell ./installer/install.ps1 --project --targets all ``` ### Verify mirrors ```bash python3 installer/verify.py ``` ### Sync mirrors from canonical source ```bash python3 installer/sync-mirrors.py ``` ## Supported targets The package is designed for: - Claude Code - OpenCode - OpenAI Codex and Codex-compatible Agent Skills environments - Google Antigravity / Gemini Agent Skills-compatible environments - other agents that discover standard `SKILL.md` skills - custom agents that can read repository instructions The installer is deliberately conservative. It never silently modifies global configuration, never downloads arbitrary remote code, and does not replace a different existing skill unless `--force` is explicitly supplied. ## Direct-build behavior The skill keeps the original structured discovery questions but skips questions already answered by the user. It then: 1. inspects the repository 2. selects the smallest suitable architecture 3. implements real frontend/backend/data/auth/integrations 4. wires real states and user journeys 5. runs lint/typecheck/tests/build 6. performs browser verification when available 7. fixes deterministic failures 8. reports actual implementation status PRD/TRD/AI-prompt documents are **not** the default deliverable. The codebase is. ## Multi-agent safety When multiple coding agents work on the same repository: - treat `skill/SKILL.md` as the source of truth - prefer project-local skill installation - keep agent mirrors identical - do not edit another agent's global configuration without explicit intent - do not commit secrets or machine-specific paths - avoid destructive git operations - do not run competing package-manager operations concurrently ## Important limitation There is no universal API that lets a GitHub repository force every arbitrary AI agent to execute an installer. Automatic installation is possible only where the host exposes compatible skill/plugin discovery or filesystem execution. This project therefore maximizes interoperability rather than making an impossible guarantee: standard skill layout + common agent discovery paths + repository instruction fallbacks + an explicit cross-platform installer. ## Original package preservation The original package was a V4 document/blueprint generator. Its detailed reference material is retained in `references/` so no useful domain knowledge is lost. The active canonical skill intentionally supersedes its document-first workflow and turns those requirements into implementation and verification rules. ## License MIT