{
  "markdown": "# Web Blueprint Engine — Universal Direct Build v7\n\nA portable, production-oriented Agent Skill for turning website and web-app requirements into **real, verified code**.\n\n## What v7 adds\n\n- One canonical `skill/SKILL.md`\n- Agent Skills-compatible layout\n- Claude Code fallback via `CLAUDE.md` + `.claude/skills/...`\n- OpenCode discovery via `.opencode/skills/...` and `.agents/skills/...`\n- Codex-compatible repository guidance via `AGENTS.md` + `.agents/skills/...`\n- Google Antigravity / Agent Skills-oriented `.agents/skills/...`\n- Claude-compatible plugin metadata in `.claude-plugin/plugin.json`\n- Cross-platform Python, shell and PowerShell installers\n- Non-destructive agent detection\n- Project-local installation by default\n- Explicit global installation\n- Safe `--force` overwrite only when intentionally requested\n- Mirror verification and synchronization utilities\n- Multi-agent repository safety rules\n- Original reference material retained under `references/`\n- Direct implementation instead of mandatory PRD/TRD/prompt generation\n\n## Repository layout\n\n```text\nskill/SKILL.md                         # canonical skill\nreferences/                            # original V4-derived detailed references\n.agents/skills/web-blueprint-engine/   # Agent Skills / Codex / Antigravity mirror\n.claude/skills/web-blueprint-engine/   # Claude Code mirror\n.opencode/skills/web-blueprint-engine/ # OpenCode mirror\n.claude-plugin/plugin.json             # Claude-compatible plugin metadata\nAGENTS.md                              # generic/Codex repository fallback\nCLAUDE.md                              # Claude Code repository fallback\ninstaller/                             # detection, install, sync, verify\nagent-adapters/                        # machine-readable interoperability policy\nskill-manifest.json                    # package metadata\n```\n\n## Install for one or many agents\n\n### Automatic project-local detection\n\n```bash\npython3 installer/install.py --project --targets auto\n```\n\n### Install all project-local adapters\n\n```bash\npython3 installer/install.py --project --targets all\n```\n\n### Global installation (explicit)\n\n```bash\npython3 installer/install.py --global --targets all\n```\n\n### Detect the host\n\n```bash\npython3 installer/install.py --detect\n```\n\n### Windows PowerShell\n\n```powershell\n./installer/install.ps1 --project --targets all\n```\n\n### Verify mirrors\n\n```bash\npython3 installer/verify.py\n```\n\n### Sync mirrors from canonical source\n\n```bash\npython3 installer/sync-mirrors.py\n```\n\n## Supported targets\n\nThe package is designed for:\n\n- Claude Code\n- OpenCode\n- OpenAI Codex and Codex-compatible Agent Skills environments\n- Google Antigravity / Gemini Agent Skills-compatible environments\n- other agents that discover standard `SKILL.md` skills\n- custom agents that can read repository instructions\n\nThe 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.\n\n## Direct-build behavior\n\nThe skill keeps the original structured discovery questions but skips questions already answered by the user. It then:\n\n1. inspects the repository\n2. selects the smallest suitable architecture\n3. implements real frontend/backend/data/auth/integrations\n4. wires real states and user journeys\n5. runs lint/typecheck/tests/build\n6. performs browser verification when available\n7. fixes deterministic failures\n8. reports actual implementation status\n\nPRD/TRD/AI-prompt documents are **not** the default deliverable. The codebase is.\n\n## Multi-agent safety\n\nWhen multiple coding agents work on the same repository:\n\n- treat `skill/SKILL.md` as the source of truth\n- prefer project-local skill installation\n- keep agent mirrors identical\n- do not edit another agent's global configuration without explicit intent\n- do not commit secrets or machine-specific paths\n- avoid destructive git operations\n- do not run competing package-manager operations concurrently\n\n## Important limitation\n\nThere 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.\n\nThis 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.\n\n## Original package preservation\n\nThe 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.\n\n## License\n\nMIT\n",
  "bytes": 4883,
  "sha": "a9ed30e22ad80720dcb8361ae5c85e89fa6126cb26d9d73420917ff1b4bdbea4",
  "repo_slug": "rishav7324/web-builder",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_rishav7324_web_builder_1f5a26fe/readme"
}