0xklen/money-machine

MONEY MACHINE

A thousand skills for AI agents that build money machines — pricing, payments, revenue ops, and the crypto rails underneath.

   

1040 skills · 444,073 words · 52 areas · 1040/1040 valid, 0 failing, 0 duplicate

What this is

1040 files, each one a procedure an agent can follow without asking questions. Every skill started as something that went wrong first: a database deleted from the wrong path, a stale cache replaying deleted file paths, a test that passed because all() over an empty list is True.

Not tips. Not prompt tricks. Commands, thresholds, and the check that proves the work happened.

How to run it

Nothing to build, no dependencies, no package manager. The skills are markdown; the tools are plain Python 3 with no third-party imports.

1. Get it

git clone https://github.com/0xklen/money-machine
cd money-machine

2. Install the skills into your agent

cp -r skills/* ~/.hermes/skills/ # Hermes Agent
cp -r skills/* ~/.claude/skills/ # Claude Code
cp -r skills/* ~/.codex/skills/ # Codex

# or point any agent at the directory directly — it is just files

3. Check it landed

ls ~/.hermes/skills | wc -l # expect 1040
python3 tools/validate.py # expect 1040/1040 valid, 0 failing, 0 duplicate

4. Use one. You do not read a skill to find it — the description is the trigger. When a task matches, read the whole file and follow ## Procedure, watch for ## Pitfalls, then run ## Verification and report the result.

5. Add one of your own

mkdir -p skills/my-new-skill
cp TEMPLATE.md skills/my-new-skill/SKILL.md # fill it in
python3 tools/validate.py --only my-new-skill # must pass, not "looks fine"
python3 tools/build-readme.py # refresh the figures above

6. Regenerate the art (optional — the pixel animations are generated, not drawn by hand)

python3 tools/pixelart.py

What is in the box

money-machine/
├── skills/ one directory per skill, 1040 of them
│ └── /
│ └── SKILL.md the skill itself — Procedure, Pitfalls, Verification
├── tools/
│ ├── validate.py the gate: structure, word floors, duplicates
│ ├── build-readme.py fills this page's figures from disk
│ └── pixelart.py generates the pixel art
├── assets/ logo and generated art
├── TEMPLATE.md the shape every skill must follow
├── BRIEF.md the 52 areas, and what belongs in each
├── LICENSE MIT
└── README.md generated — edit the template, not this file

The fifty-two areas

agent-01 Agent self-management · agent-02 Agent honesty and evidence · software-01 Correctness under change · software-02 Running systems · software-03 Delivery and supply chain · crypto-01 Keys and signing · crypto-02 Contract safety · crypto-03 DeFi and markets · crypto-04 On-chain interaction · judgment Research and evaluation · agent-03 Memory and knowledge · agent-04 Multi-agent orchestration · agent-05 Tool and API integration · agent-06 Failure containment · agent-07 Communication · sec-01 Security review · sec-02 Privacy and data · sec-03 Adversarial robustness · data-01 Pipelines · data-02 Analytics · infra-01 Containers · infra-02 Networking · infra-03 Storage · infra-04 IaC and environments · perf-01 Web performance · perf-02 Backend performance · prod-01 Product and requirements · prod-02 Design engineering · crypto-05 Wallets and custody · crypto-06 MEV and orderflow · crypto-07 Rollups and L2 ops · crypto-08 Stablecoins and payments · crypto-09 NFT and on-chain art · crypto-10 Governance and DAOs · crypto-11 Data and indexers · crypto-12 Compliance-aware work · ai-01 Model selection and prompting · ai-02 Evaluation and red-teaming · ai-03 RAG and knowledge · ai-04 Agents and tool use · craft-01 Docs and writing · craft-02 Code review · craft-03 Debugging method · craft-04 Incident command · craft-05 Estimation and planning · craft-06 Negotiation and scope · craft-07 Teaching and onboarding · craft-08 Automation design · craft-09 Time and calendar · craft-10 Money and units · money-01 Pricing and unit economics · money-02 Revenue operations

What this is not

- Not policy or a safety layer. These are working procedures, not guardrails. - Not filler. The validator rejects stubs and duplicates, so a thin area gets fixed or cut rather than padded to hit a number. - Not advice about your jurisdiction, your security model, or your money. Several skills instruct the reader to stop and ask a human. That is the point of them. - Not finished. Skills go stale the moment the world moves; skill-hygiene and prune-stale-facts exist because keeping this honest is ongoing work.

Four to read first

reconcile-state-before-acting what is actually running, cached and listening
poison-a-fixture-to-prove-a-detector prove the detector catches a planted fault
simulate-a-transaction-before-broadcast never discover a revert on mainnet
read-back-state-after-deploy a successful call is not a successful task

Contributing

Bring the failure, not the topic. A skill earns its place by describing something that actually went wrong, the procedure that prevents it, and how to prove the procedure ran. Then:

python3 tools/validate.py # must print no failures
python3 tools/build-readme.py # keeps the figures above true

License

MIT — see LICENSE. Use them, fork them, ship them in your own agent.

every skill in here was paid for by a mistake