Skip to content
Skill

genesis-architect

by maioio

AI Summary

Skill metadata (version, author, license, optional_mcps, fallback) lives in manifest.json and plugin.json. Claude Code only reads and from this frontmatter; extra keys are silently ignored, so they belong with

Install

Copy this and paste it into Claude Code, Cursor, or any AI assistant:

I want to install the "genesis-architect" skill in my project.

Please run this command in my terminal:
# Install skill into your project
mkdir -p .claude/skills/genesis-architect && curl --retry 3 --retry-delay 2 --retry-all-errors -o .claude/skills/genesis-architect/SKILL.md "https://raw.githubusercontent.com/maioio/genesis-architect/main/SKILL.md"

Then restart Claude Code (or reload the window in Cursor) so the skill is picked up.

Description

Use when starting a new project. Genesis finds real production failures from similar GitHub repos, then builds a working MVP immediately - not just documents. Two modes: fast-mvp (5 min research cap, builds right after) and deep-research (full analysis). Success = user can run the project. Triggers on: "genesis init [vision]", "I want to build X", "scaffold", "new project", "set up project", "start building", "create a tool", "make a CLI", "bootstrap", "בנה פרויקט", "צור פרויקט", "התחל פרויקט".

Genesis Architect

> Research fast. Build immediately. Genesis turns production failures into a working MVP. ---

Language Detection

Detect the user's language from their first message. Respond in that language throughout all phases. Default to English for unrecognized languages. ---

Invocation

When the user writes genesis init [description], extract the vision and skip Phase 1 questions. Auto-detect archetype and scale from the description. These are surfaced for confirmation in Phase 5 before any files are created - never silently assumed. Phase 0 always runs regardless of invocation method. genesis init --from-prd [file] - read a PRD file (e.g., from idea-os or similar product planning tools). Extract: project name, core purpose, target users, scale, constraints. Use as Phase 1 answers and as additional search signals in Phase 2. Skip Phase 1 questions entirely. genesis init --from-team-config - read .genesis.json from the current directory. Restore language, tier, and research context from a teammate's prior Genesis run. Skip Phases 1-5. Required fields in .genesis.json: language, tier (minimalist|scalable), vision. If any required field is missing, abort with: ".genesis.json is missing field: [field]. Run genesis init to generate it." genesis init --fast-mvp [description] - Research Budget mode. Hard limits: 5 min research, 10 repos, 30 issues, 5 Exa sources. After budget exhausted, FORCE_BUILD=TRUE: skip to Phase 6 immediately. Produces BUILD_PACKET.md instead of separate RESEARCH.md/PITFALLS.md. Announce at start: "Fast MVP mode - 5 min research cap, then building immediately." genesis init --deep-research [description] - Full research mode (default). No time cap. All streams. Detailed RESEARCH.md + PITFALLS.md + ROADMAP.md. genesis audit [path] - run Phases 2-4 on an existing codebase. Delivers PITFALLS.md and RESEARCH.md. No scaffold generated. Pre-flight Check (Phase 0.5) does not apply to genesis audit - it is an explicit command. Before Phase 2, infer vision context from the existing codebase: read README.md if present, scan package.json/pyproject.toml/go.mod for name and description. Use these as Phase 1 substitutes. If nothing useful is found, ask one question: "Describe what this project does (one sentence)." Then run Phases 2-4 directly. Phase 4's "Before proceeding to Phase 5" instruction does not apply to genesis audit - terminate after delivering PITFALLS.md and RESEARCH.md. genesis harden [path] - security and quality upgrade for an existing project (defaults to current directory). Runs a gap scan and injects missing standards: • Check for: secret-scanning workflow, SAST workflow, quality-gate config, strict .gitignore • Inject any missing files using templates from references/security-templates.md • Scan src/ for common security gaps: missing input sanitization, hardcoded strings resembling secrets (regex: [A-Za-z0-9]{32,}), unsafe file opens without path validation • Output a status table: what was injected (auto) vs what needs manual action (tokens, org name) Pre-flight Check (Phase 0.5) does not apply - this is an explicit command. Read references/architecture-patterns.md for boilerplate templates. Read references/mcp-strategy.md for MCP usage and fallback logic. ---

Phase 0: Environment Probe

Run python scripts/env_probe.py and parse the JSON. Fields: os, wsl, python_version, package_managers.{python,node}, windows_scripts_path. Store the result for use in Phases 3, 5, and 6. If the script fails (e.g. Python missing), ask once: "What OS and Python version are you on?" Convention scan: silently check nearby projects for HTTP client, test framework, DB, formatter. Present once in Phase 5: "Your projects use [X]. Match? [Y/n]" Windows PATH check: when os == "windows" and wsl == false, use windows_scripts_path from the probe. After pip install -e ., if command not recognized: session fix $env:PATH += ";[Scripts path]" (PS) or set PATH=%PATH%;[Scripts path] (CMD). When wsl == true, treat as Linux - skip Windows PATH fixes. ---

Discussion

0/2000
Loading comments...

Health Signals

MaintenanceCommitted 16d ago
Active
AdoptionUnder 100 stars
44 ★ · Niche
DocsREADME + description
Well-documented

GitHub Signals

Stars44
Forks6
Issues8
Updated16d ago
View on GitHub
MIT License

My Fox Den

Community Rating

Sign in to rate this booster

Works With

Claude Code