SoloForge

Full-lifecycle SOP system for Claude Code — from idea validation to development to marketing to growth

SoloForge turns Claude Code from a code generator into an autonomous business operating system. It enforces quality gates across the entire lifecycle — business strategy, development, marketing, growth, and operations — through automated SOP skills, safety hooks, and a critical thinking protocol that forces adversarial reasoning before conclusions.

---

The Problem

AI coding agents don't check their own output. They claim work is done without testing, rationalize away quality issues ("pre-existing", "out of scope", "follow-up task"), and produce inconsistent results across sessions. Beyond code, they lack any structured approach to business decisions, marketing, or growth.

SoloForge solves this with enforced quality gates at every workflow stage -- a layered skill and hook system where each layer reinforces the others, covering the full journey from idea to revenue.

---

Philosophy

• Adversarial reasoning: AI should argue both sides before concluding. The thinking protocol requires searching the web, finding contradictions, and synthesizing -- not just pattern-matching from training data.
• Quality through process, not filtering: Enforcement happens through skill SOPs at each workflow stage, not post-hoc output filtering. Every action has a checklist; every checklist has verification steps.
• Critical thinking is structural: Epistemic discipline, verification over assumption, and quality intolerance are built into the CLAUDE.md principles that sit in context permanently -- not bolted on as afterthoughts.

---

How It Works

SoloForge is a 3-layer architecture where each layer reinforces the others:

Layer 1: CLAUDE.md  (133 lines)  -- Principles always in context
Layer 2: Skills     (26 SOPs)    -- Auto-triggered at workflow stages, loaded on-demand
Layer 3: Hooks      (3 hooks)    -- Deterministic enforcement, runs outside the LLM

• Layer 1 sets the ground rules: epistemic discipline, verification over assumption, quality intolerance.
• Layer 2 provides step-by-step SOPs that Claude invokes before each action (coding, committing, reviewing, pricing, marketing, etc.).
• Layer 3 runs outside the model entirely -- shell scripts that block dangerous operations and flag common mistakes in edited files.

---

The Workflow

flowchart TD
    A[biz-think] -->|Acid test pass| B[brand-build]
    A -->|Acid test fail| A2[biz-validate]
    A2 -->|Validated| B
    B --> C[project-init]
    C --> D[pre-code]
    D --> E[Code + self-review]
    E --> F[test-gate]
    F --> G[pre-commit]
    G --> H[pre-pr]
    H --> I{UI changes?}
    I -->|Yes| J[ux-audit]
    I -->|No| K[Merge]
    J --> K
    K --> L[post-merge]
    L --> M[product-eval]
    M -->|Score >= 70| N[copy-craft + content-create]
    M -->|Score < 70| M
    N --> O[content-distribute]
    O --> P[community-engage + sales-close]
    P --> Q[growth-track]
    Q -->|Next feature| D

    style A fill:#4a9eff,color:#fff
    style M fill:#4a9eff,color:#fff
    style Q fill:#2ecc71,color:#fff

Loading

Every stage has a skill. Every skill has a checklist. Skipping a stage means skipping its checks -- and the principles in CLAUDE.md push back against that.

Skill Chain

Every skill ends with Next Steps linking to related skills. This creates a natural chain: biz-think points to brand-build and project-init, post-merge points to product-eval, product-eval points to copy-craft and content-create, and so on. You never have to remember what comes next -- each skill tells you.

---

Quick Install

git clone https://github.com/Docat0209/SoloForge.git
cd SoloForge && ./install.sh

The installer copies skills, hooks, and the CLAUDE.md template into your ~/.claude/ directory.

To install into an existing project manually:

# Copy the CLAUDE.md template
cp claude.md ~/.claude/CLAUDE.md

# Copy hooks
cp hooks/safety-guard.sh ~/.claude/hooks/
cp hooks/post-edit-review.sh ~/.claude/hooks/
cp hooks/skill-eval.sh ~/.claude/hooks/

# Copy skills
cp -r skills/* ~/.claude/skills/

# Copy hook configuration
cp settings/hooks.json ~/.claude/settings.json

---

What's Included

Skills (26 SOPs)

Development Workflow (10)

Skill	What It Does
pre-code	Issue + branch + SOLID design check before coding
issue-create	Enforces issue templates (feat/fix/chore)
project-init	Git + CI/CD + test runner + branch protection
self-review	6-step code review (Google, Clean Code, OWASP)
test-gate	3-layer testing (Fowler Pyramid, AAA pattern)
pre-commit	Atomic commits (Chris Beams, Conventional Commits)
pre-pr	PR quality ≤200 lines (Google Small CLs)
ux-audit	Visual + UX + Performance audit via Playwright
post-merge	4-step verification after merge
search-eval	Source credibility + dependency evaluation

Business & Strategy (5)

Skill	What It Does
biz-think	5 viability tests + success decomposition + zero-audience playbook
biz-validate	Hypothesis → experiment → learn loop (Running Lean + Pretotyping)
product-eval	4-dimension scoring /100 (Nielsen, HEART) + auto-iteration if <70
roadmap-steer	RICE scoring, feature prioritization, saying no
finance-ops	Value-based pricing, pricing psychology, MRR tracking, tax basics

Marketing & Growth (7)

Skill	What It Does
brand-build	Positioning, voice definition, Build in Public, tribe identity
content-create	Platform-optimized articles from customer perspective (They Ask You Answer)
content-distribute	One piece → 15+ platform-adapted versions
community-engage	Watering holes strategy, value-first engagement
copy-craft	Conversion copy: PAS/AIDA, Cialdini persuasion, landing pages
sales-close	Dream 100, cold email, SPIN selling, pipeline management
growth-track	AARRR funnel, North Star metric, budget, partnerships

Operations (4)

Skill	What It Does
support-ops	P0-P3 triage, churn prevention, onboarding sequences
infra-ops	Monitoring, 3-2-1 backups, incident response, security
data-decide	Event taxonomy, cohort analysis, A/B testing, decision framework
legal-guard	ToS, Privacy Policy, GDPR/CCPA, cookie consent

Hooks (3 guards)

Hook	Event	What It Does
skill-eval.sh	UserPromptSubmit	Lightweight skill gate reminder (~50 tokens) — forces evaluate → activate → work sequence
safety-guard.sh	PreToolUse (Bash)	Blocks: --force push, reset --hard, checkout ., clean -f, branch -D, --no-verify, credential dir access
post-edit-review.sh	PostToolUse (Write/Edit)	Detects: debug statements, TODO/FIXME comments, hardcoded secrets, Bearer tokens, .env modifications

---

Recommended Agents

SoloForge's agent prompts are built on top of Agency Agents -- a collection of 100+ specialized agent definitions for Claude Code. Install them to get the full roster of sub-agents (Backend Architect, Frontend Developer, UX Researcher, Growth Hacker, etc.) that the orchestrator delegates to.

git clone https://github.com/msitarzewski/agency-agents.git
cp -r agency-agents/agents/* ~/.claude/agents/

---

Recommended MCP Servers

These optional MCP servers enhance specific skills:

Server	Purpose	Used By	Install
Sequential Thinking	Structured multi-step reasoning	biz-think, product-eval	claude mcp add sequential-thinking -- npx -y @modelcontextprotocol/server-sequential-thinking
Playwright	Browser automation for visual review	ux-audit, post-merge	claude mcp add playwright -- npx @playwright/mcp@latest
Chrome DevTools	Browser debugging, console errors, performance	post-merge, product-eval	claude mcp add chrome-devtools -- npx chrome-devtools-mcp@latest
Figma	Design ↔ code bidirectional (Figma → code, code → Figma)	All frontend work	claude mcp add --transport http figma https://mcp.figma.com/mcp
TypeScript LSP	Auto type-check after every edit, catch errors before build	All TypeScript projects	claude plugin install typescript-lsp

---

Customization

Add a new skill

1. Create a directory under skills/ with your skill name:

   mkdir skills/my-skill

2. Add a SKILL.md file with the SOP checklist. Follow the pattern of existing skills.
3. Register the trigger in claude.md under the Skill Gate Protocol section:

   - Before [action] → invoke `my-skill`
   

Modify CLAUDE.md

Edit claude.md directly. The 133-line template is intentionally concise -- every line earns its place in the context window. Add principles sparingly.

Add a new hook

1. Write a shell script in hooks/. Use exit code 2 to block (PreToolUse) or 0 to allow.
2. Register it in settings/hooks.json under the appropriate event (PreToolUse or PostToolUse).

---

Uninstall

./uninstall.sh

This removes all SoloForge files from ~/.claude/ and restores your previous configuration if a backup exists.

---

Credits

SoloForge synthesizes practices from established engineering, product, and business frameworks:

• Google Engineering Practices -- code review standards, small CLs
• Clean Code (Robert C. Martin) -- self-review principles
• SOLID Principles -- design constraints in pre-code
• NASA Power of 10 -- pre-commit verification rules
• OWASP Top 10 -- security checks in self-review
• Chris Beams on Git Commits -- commit message format
• Conventional Commits -- commit type enforcement
• Martin Fowler Testing Pyramid -- 3-layer test strategy
• Nielsen Usability Heuristics -- product evaluation dimensions
• Google HEART Framework -- metrics in product-eval
• Cognitive Walkthrough -- FTUE evaluation method
• Porter's Value Chain -- business analysis in biz-think
• Jobs to Be Done (Clayton Christensen) -- customer lens
• Lean Canvas (Ash Maurya) -- business model validation
• AARRR Pirate Metrics -- growth funnel analysis
• Sean Ellis PMF Survey -- product-market fit signal
• They Ask, You Answer (Marcus Sheridan) -- content creation
• SPIN Selling (Neil Rackham) -- sales methodology
• Dream 100 (Chet Holmes) -- outbound sales
• Influence (Robert Cialdini) -- persuasion in copy and pricing
• Running Lean (Ash Maurya) -- business validation loop
• Pretotyping (Alberto Savoia) -- hypothesis testing
• Intercom on Product Management -- RICE scoring
• Tribes (Seth Godin) -- community and brand
• The Mom Test (Rob Fitzpatrick) -- customer interview validation

---

License

MIT -- Copyright (c) 2026 Shane Ho