// Solo Factory + Solograph

Agent-driven startup development in flow

Two open-source tools that turn Claude Code into a full startup factory with memory and skills.

solo-factory → solograph → PRDs built with this →

// install

Skills (any agent)
npx skills add fortunto2/solo-factory --all
MCP server (Claude Code)
claude mcp add -s project solograph -- uvx solograph
Plugin (everything together)
claude plugin marketplace add fortunto2/solo-factory
claude plugin install solo@solo --scope user
Or install solograph as a Python package
uv install solograph

// Solograph

MCP server with 11 tools for Claude Code

Code and knowledge graph on FalkorDB. Semantic vector search over code and docs. Memory of past Claude Code sessions. Knowledge base search. Web search via SearXNG (optional).

// architecture

> 100% local. FalkorDB embedded, no Docker, no cloud
> MLX embeddings on Apple Silicon (~1ms), fallback for other platforms
> Graph nodes: Project, File, Symbol, Package, Session
> Graph edges: IMPORTS, CALLS, INHERITS, DEPENDS_ON, EDITED

// 11 tools

codegraph_query Cypher queries over code graph: files, symbols, dependencies, inheritance
codegraph_explain Project architecture overview: stack, layers, patterns, key files
codegraph_stats Graph statistics: project, file, symbol, and package counts
codegraph_shared Shared dependencies across projects
project_code_search Semantic search over project code and docs (auto-indexed)
project_info Project info from registry: stack, path, dependencies
kb_search Semantic search over knowledge base (markdown + YAML frontmatter)
session_search Search past Claude Code sessions ("how did I solve X?")
source_search Search YouTube transcripts, Telegram, and other indexed sources
source_tags Auto-detected topics with video counts
web_search Web search via SearXNG (optional, if configured)

// Solo Factory

Plugin with 10 core skills for Claude Code

Each skill is a carefully crafted prompt + instructions that turns Claude Code into a specialist for one stage of the startup pipeline. Skills use SoloGraph MCP tools, web search, and code intelligence. 

idea → /solo:research → /solo:validate → /solo:scaffold → /solo:setup → /solo:plan → /solo:build → shipped

// core skills (10)

/solo:init First Setup

Personalized founder setup. Org defaults, manifest, dev principles, POTOK calibration, stack selection. Creates ~/.solo-factory/defaults.yaml (org-level) and .solo/ in project (philosophy, principles, stacks). Asks guided questions across Philosophy, Dev Preferences, and Decision Style.

/solo:research Market Research

Competitors, SEO/ASO, naming, domains, market sizing. Hybrid search: Claude WebSearch + SearXNG. Reddit, YouTube, Product Hunt, G2, Crunchbase. Domain triple-verification (whois + dig + RDAP). Generates research.md with GO/NO-GO/PIVOT recommendation.

/solo:validate Idea Validation

Score idea through POTOK 6-layer framework + S.E.E.D. niche check (Searchability, Evidence, Ease, Demand). Auto-selects stack from 9 templates. Generates scored PRD with problem, ICP, features, KPIs, risks, pricing, launch strategy. 5 minutes from idea to verdict.

/solo:scaffold PRD to Project

PRD to running project in 2 minutes. Creates directory structure, CLAUDE.md, Makefile, README, configs. Git init + GitHub repo. Studies existing projects via SoloGraph for patterns. Uses Context7 for latest library versions.

/solo:setup Dev Workflow

Auto-generate docs/workflow.md from existing PRD + CLAUDE.md. TDD policy, commit strategy, verification checkpoints. Zero questions -- reads your codebase and decides.

/solo:plan Implementation Plan

Explore codebase, write spec.md + plan.md with phased tasks. Uses SoloGraph MCP: session_search, project_code_search, codegraph_query. Auto-classifies track type (bug/refactor/chore/feature). Zero interactive questions.

/solo:build Ship It

Execute tasks from plan.md with TDD workflow (Red-Green-Refactor). Auto-commits per task with conventional commits. Phase gates between implementation phases. Handles errors, test failures, and resumption.

/solo:swarm Parallel Research [3 agents]

3 agents in parallel: Market Researcher (competitors, pricing), User Researcher (Reddit/HN pain points), Technical Analyst (feasibility, GitHub solutions). Synthesizes into research.md with GO/NO-GO recommendation.

/solo:potok Decision Filter

Run any decision through POTOK 6-layer filter. Scores each layer 1-10, weighted overall score. Verdict: GO (>7) / PAUSE (5-7) / NO-GO (<5). Saves decision record to docs/decisions/.

/solo:audit KB Health Check

Audit all .md files: frontmatter validation, broken internal links, tag consistency, orphaned files, stale drafts, coverage gaps. Detailed report with fix recommendations.

// content & growth skills (7)

Post-launch: content, SEO, metrics, community outreach.

/solo:content-gen Content Pack

Generate LinkedIn, Reddit, Twitter/X posts from your PRD. 3 ready-to-publish pieces with messaging extracted from product specs.

/solo:seo-audit SEO Audit

Health check for any URL: meta tags, OG, JSON-LD, sitemap, robots.txt, SERP positions. Score 0-100 with fix recommendations.

/solo:landing-gen Landing Page

Generate hero, features, CTA, SEO meta, A/B headlines. Scaffolds actual Astro/Next.js page components if project detected.

/solo:video-promo Video Script

Promo video plan: script, storyboard, shot list. Detects Remotion/Life2Film in project and outputs compatible config.

/solo:community-outreach Community Outreach

Find relevant Reddit/HN/ProductHunt threads. Draft value-first responses (not spam). ProductHunt launch checklist included.

/solo:metrics-track Metrics Setup

PostHog funnel config, KPI benchmarks, kill/iterate/scale decision thresholds. Ready-to-implement tracking plan.

/solo:index-youtube YouTube Indexer

Index YouTube channels into FalkorDB source graph via solograph CLI. Powers the research layer with transcript search.

// agents (3)

Specialized subagents. Used by /solo:swarm and available for direct spawning.

researcher Sonnet

Market research specialist. Checks existing knowledge first, searches multiple sources, cites everything, quantifies claims. Outputs competitor table, pain points, market sizing, GO/NO-GO recommendation.

code-analyst Haiku

Code intelligence analyst. Uses codegraph for dependency analysis, architecture review, cross-project patterns. Finds existing implementations before writing new code. Session archaeology to recover past solutions.

idea-validator Sonnet

Validation specialist. Runs POTOK 6-layer analysis, checks manifesto alignment, selects tech stack, generates scored PRD. Outputs opportunity score 0-10 with next action.

// stack templates (9)

Production-ready configs used by /solo:scaffold and /solo:validate. Includes tooling, patterns, architecture conventions.

Stack Platform
nextjs-supabase Web
nextjs-ai-agents Web + AI
astro-static Web
astro-hybrid Web
ios-swift iOS
kotlin-android Android
python-api Server
python-ml ML/CLI
cloudflare-workers Edge

// decision frameworks

POTOK -- 6-layer decision filter

Every idea passes through 6 layers. Each scored 1-10 with weighted total. GO (>7) / PAUSE (5-7) / NO-GO (<5).

Perception -- circle of competence
Origin -- timing, Lindy, reversibility
Trail -- minimum viable action
Openness -- risk/reward asymmetry
Kinship -- network effects, moat
*meta -- mortality filter

S.E.E.D. -- niche validation check

Quick viability check with hard kill flags. If any kill flag triggers, the idea is dead.

Searchability -- SERP analysis, can you rank?
Evidence -- Reddit/forum signals, real pain?
Ease -- MVP in days, not months?
Demand -- keywords, willingness to pay?

// example: idea to PRD in 5 minutes

1. Research the market
$ claude "/solo:research AI tools directory for dentists"

→ research.md: competitors, Reddit pain points, keywords, domains

2. Validate and score
$ claude "/solo:validate dental AI directory"

→ POTOK score 6.5/10, S.E.E.D. pass, PRD with stack selection

3. Scaffold the project
$ claude "/solo:scaffold"

→ Astro project with CLAUDE.md, Makefile, configs, GitHub repo

4. Plan and build
$ claude "/solo:plan" && claude "/solo:build"

→ spec + plan → TDD → auto-commits → deployed

// principles

Every PRD, CLAUDE.md, and decision follows these principles (customizable via /solo:init):

>AI as foundation, not feature
>Privacy-first, offline-first
>One pain, one feature, ship in days
>Speed over perfection
>Antifragile architecture
>Revenue is feedback, not the goal
GitHub: solo-factory → GitHub: solograph → PRD catalog →
← back to terminal