The Claude Code config a real solo founder uses every day to run a 1-person company.
11 agents · 10 skills · 10 commands · 4 hooks · 6 months of real workflow. Drop into
~/.claude/· ready in 60 seconds.
A complete operating system for solo founders using Claude Code. Not another awesome-claude-code link list — a working configuration that's been refining itself for 6 months while running real revenue products.
| Agent | What it does |
|---|---|
architect |
Tech stack decisions · data flow · API contracts · no code |
planner |
Idea → executable phased plan with validation gates |
code-reviewer |
Security · quality · performance check before merge |
security-reviewer |
Pre-deploy security audit |
test-engineer |
Test plan + coverage gap analysis |
frontend-expert |
UI / UX / Tailwind / Next.js specialist |
backend-expert |
API / DB / queue / observability |
doc-writer |
README / changelog / customer docs |
customer-support |
Triage messages · draft replies |
marketing |
Content · SEO · case studies · proposals |
validator |
Cross-check claims · catch hallucinations |
| Skill | When to use |
|---|---|
idea-eval |
New product idea? Score on 6 dims · 22+ = green light |
lead-intake |
Discovery call → score the lead 1-10 |
proposal-gen |
Lead → SOW + pricing draft |
case-study-gen |
Delivered project → 5 reusable content pieces |
prior-art-search |
Before building · OSS-First search same problem |
eval-refine |
Output not good enough? Eval-driven iteration |
ship-checklist |
Pre-release safety net · what could break |
release-window |
When NOT to ship (high-risk windows) |
soft-launch |
Test with 5 real users before public |
content-blitz |
One topic → 5 platforms in 2 hours |
| Command | Purpose |
|---|---|
/plan |
Spawn planner · output executable plan |
/eval |
Score idea on 6 dims |
/save-session |
Save handoff memo (real handoffs only) |
/resume-session |
Reload last session context |
/learn |
Capture method/rule/anti-pattern |
/learn-review |
Review learning candidates |
/orchestrate |
Multi-agent task coordination |
/loop |
Auto-iterate task with self-pacing |
/skill-create |
Create new reusable skill |
/cost |
Today's API spend snapshot |
| Hook | Triggers when | Does |
|---|---|---|
session-start.js |
Session opens | Display state board (date · last session · pipeline · assets · cost) |
pre-bash-safety-check.js |
Bash tool runs | Block rm -rf / · warn git push --force |
pre-compact-reminder.js |
Context auto-compact | Suggest /save-session if real handoff |
stop-learning-reminder.js |
Session ends | Conditional /save-session + /learn ping |
📖 See how these pieces stitch into real workflows → RECIPES.md
5 actual sequences I run: idea-to-ship in 7 days · discovery-call to signed contract · delivery to 5 reusable content pieces · bug-report to release · multi-Claude without collisions.
After 6 months of running a 1-person company with Claude Code, I built up:
- Patterns that avoided wasting weeks on bad ideas
- Patterns that kept me alive when API costs spiked
- Patterns that caught hallucinated facts before they shipped to clients
- A real operating rhythm where Claude does 80% of the routine work
Most awesome-claude-code lists give you 50 tools to evaluate. This gives you the working set — what's actually battle-tested in real revenue work.
- ❌ Toy agents that look impressive but don't get used
- ❌ "Universal" templates that don't fit any real workflow
- ❌ My private business strategy / customer list / pricing
- ❌ Specific to one stack (works with any stack you choose)
Everything personal has been stripped. Everything that ships is what I actually use.
Most Claude Code configs are collections of agents. This pack ships 4 disciplines that don't exist elsewhere — built from real production pain:
The pack itself ships with scripts/sanitize.sh patterns that strip private business terms. If you fork it for your own stack, the pattern is reusable: never push code that mentions a customer or unreleased product by mistake.
When two Claude sessions work in parallel, they can't see each other's chat — only files. The CLAUDE.md template ships a docs/contracts/ discipline: cross-module interface specs that prevent two sessions from overwriting each other's work. Battle-tested after a real collision.
Any agent calling a paid API must show a dry-run cost estimate first, wait for explicit --approve, then execute. Stops the "$200 spent in one bug-hunting session" failure mode. The discipline is encoded in CLAUDE.md and enforced by pre-bash-safety-check.js.
Any "current state" file (pipeline status, asset board) carries a TTL. When Claude reads an expired state file, it actively warns the founder instead of pretending the data is fresh. Prevents AI from giving decisions based on a 3-month-old snapshot.
These four are the reason the pack feels different on day 7 than day 1 — they compound.
📁 See concrete examples of what your
docs/looks like after running this for 1-3 months → examples/. Filled CLAUDE.md, populated pipeline, real ADR, contract specs, session memos.
Honest positioning — there are bigger packs out there. Here's where each one wins.
| Affaan ECC (176k ⭐) | wshobson/agents (35k ⭐) | claude-code-pack (this repo) | |
|---|---|---|---|
| Lens | Agent harness performance optimization | Subagent + skill collection | Solo founder operating system |
| Audience | Dev teams optimizing Claude Code workflows | Generalist Claude Code users | 1-person companies / indie hackers |
| Philosophy | More & deeper | More & broader (185 agents · 153 skills) | Curated · workflow-stitched · less = more |
| Plugin install | ✅ Marketplace native | ✅ Marketplace native | ✅ Plugin manifest included |
| Demo videos | ✅ Yes | ✅ Yes | ❌ Not yet |
| Multi-Claude collab discipline | partial | ❌ | ✅ Contracts pattern |
| API cost approval gates | ❌ | ❌ | ✅ Encoded in CLAUDE.md + hooks |
State file decay (staleAfterDays) |
❌ | ❌ | ✅ Mandatory for rolling state |
| Real workflow recipes | ❌ | ❌ | ✅ RECIPES.md — 5 sequences |
| Sample running-state artifacts | ❌ | ❌ | ✅ examples/ folder |
| Free OSS use | ✅ Fully free | ✅ Fully free | ✅ Free under MIT/CC-BY-NC |
| Paid tier | ❌ | ❌ | Optional · for 1:1 setup + private updates |
When to pick this one over the others:
- You're a solo founder / indie hacker (not a 50-person team)
- You want curated workflows · not a 200-agent buffet
- You care about API cost discipline + multi-Claude safety as much as agent count
When to pick Affaan / wshobson over this:
- You want the absolute biggest collection
- You're a Claude Code power user wanting every possible tool
- You don't need solo-founder-specific patterns
The three packs aren't substitutes — they're different tools for different operators.
# 1. Download
git clone https://github.com/pagoda111king/claude-code-pack.git
cd claude-code-pack
# 2. Backup your existing .claude (if any)
cp -r ~/.claude ~/.claude.bak.$(date +%s) 2>/dev/null || true
# 3. Install
cp -r template/.claude/* ~/.claude/
# 4. Customize CLAUDE.md (this is your operating manual · read it once)
$EDITOR ~/.claude/CLAUDE.md
# 5. Test it
# Open Claude Code · type "/cost" → if you see today's spend · you're liveSee INSTALL.md for advanced setup (per-project overlays · custom hooks · CI integration).
The full pack is open source under the licenses below. You can read it, fork it, use it in your own work — no payment needed.
The paid tiers are for people who want a relationship, not just files:
| Tier | Price | What's actually different |
|---|---|---|
| Starter | $39 | Same files as the public repo · packaged zip + email when v1.1 ships |
| Pro ⭐ | $79 | + 30 use-case walkthroughs (when to fire which agent · real examples) + private channel for Q&A |
| Founder | $199 | + 1-hour 1:1 setup call · I customize CLAUDE.md for your specific stack and workflow |
Why $79? Roughly 1 hour of senior dev billable time. This pack distills 40-60 hours of trial-and-error figuring out how to make Claude Code feel like a co-founder instead of a fancy autocomplete.
If you just want the config — git clone and you're done. If you want my time → paid tier.
MIT for the code · CC-BY-NC for the docs/skills/agents (no resale of derivative works).
You can:
- ✅ Use in your own projects
- ✅ Modify for your team's needs
- ✅ Share modifications with your team
You can't:
- ❌ Resell as a competing product
- ❌ Strip attribution and republish
Q: Will this work with Cursor / Aider / Continue? A: It's designed for Claude Code. The agent definitions and hooks use Claude Code-specific schemas. Skills and CLAUDE.md patterns transfer broadly.
Q: I'm not a "solo founder" — does this still help? A: Yes if you're a solo dev / indie hacker / consultant / freelancer. Anyone running their own workflow. Less useful if you're inside a 50-person team with shared infra.
Q: How often is it updated? A: Real updates as I learn things. Buyers get free updates for 12 months.
Q: Refund policy? A: 14 days · no questions · email founder@example.com.
Built by pagoda111king — solo founder running a 1-person company, shipping real revenue products with Claude Code as the daily driver. This pack is what 6 months of that work compressed into a config bundle.
If something here saved you a week, that's the whole point.