Coding craft skills for AI agents — 7 discipline tracks, 21 SKILL.md files.
agent-coda is a collection of engineering craft skills in OpenClaw SKILL.md format, organized into seven discipline tracks. Each skill is a structured, substantive guide an AI agent can load and follow — not documentation, not a tutorial, but operational craft that shapes how an agent codes, deploys, secures, and maintains systems.
These are skills, not opinions. Every pattern here reflects what works in production and what reliably fails.
| Track | Description | Key Skills |
|---|---|---|
🛠️ app |
Application engineering discipline | Karpathy principles, code review, refactoring, test strategy |
🚀 devops |
Delivery pipeline and deployment craft | Pipeline discipline, deployment safety, container craft |
🔐 secops |
Security operations and hardening | Threat modeling, secrets hygiene, vuln triage |
☁️ infra |
Infrastructure as code and cloud discipline | IaC discipline, cost governance, provisioning patterns |
🖥️ sysadmin |
System administration and operational hygiene | OS hardening, shell discipline, service management |
📊 data |
Data pipelines and ML operations | Pipeline patterns, MLOps |
⛓️ web3 |
Blockchain engineering and on-chain patterns | Smart contract craft, on-chain patterns |
| Skill | One Line | |
|---|---|---|
| 🧠 | app/karpathy-principles |
Think before coding. Simplicity first. Surgical changes. Verify goals. |
| 👁️ | app/code-review |
What to find in a review — correctness, safety, clarity, in that order. |
| 🔧 | app/refactoring |
When to refactor, how to scope it, and when to stop. |
| 🧪 | app/test-strategy |
Right tests for the right layer — unit, integration, e2e, and fuzz. |
| Skill | One Line | |
|---|---|---|
| 🔁 | devops/pipeline-discipline |
CI/CD pipelines that gate quality, not just run commands. |
| 🛟 | devops/deployment-safety |
Canary, blue/green, rollback — never deploy without an exit. |
| 📦 | devops/container-craft |
Lean, reproducible, secure containers. |
| Skill | One Line | |
|---|---|---|
| 🎯 | secops/threat-modeling |
Define what you're protecting before you write the first line. |
| 🔑 | secops/secrets-hygiene |
Secrets belong in vaults, not in code, logs, or environment files. |
| 🩹 | secops/vuln-triage |
How to assess, prioritize, and remediate security findings. |
| Skill | One Line | |
|---|---|---|
| 📋 | infra/iac-discipline |
Infrastructure is code. Review it, version it, test it. |
| 💰 | infra/cloud-cost-governance |
Cost is a feature. Budget before you provision. |
| 🏗️ | infra/provisioning-patterns |
Repeatable, auditable provisioning — not snowflake servers. |
| Skill | One Line | |
|---|---|---|
| 🔒 | sysadmin/os-hardening |
Minimal attack surface from day one — not after the incident. |
| 💻 | sysadmin/shell-discipline |
Safe, readable, maintainable shell scripts. |
| ⚙️ | sysadmin/service-management |
systemd, process supervision, and graceful failure. |
| Skill | One Line | |
|---|---|---|
| 🔁 | data/pipeline-patterns |
Idempotent, partitioned, observable data pipelines. |
| 🤖 | data/mlops |
From notebook to production: experiment tracking, serving, and monitoring. |
| Skill | One Line | |
|---|---|---|
| 📜 | web3/smart-contract-craft |
Security patterns, upgrade strategies, and audit readiness for on-chain code. |
| ⛓️ | web3/onchain-patterns |
Event indexing, reorg handling, and on-chain/off-chain bridge discipline. |
cd <your-skills-directory>
git clone https://github.com/prapanch/agent-coda.gitgit clone https://github.com/prapanch/agent-coda.git /path/to/agent-coda
ln -s /path/to/agent-coda <your-skills-directory>/agent-coda# Example for OpenClaw:
openclaw skills install prapanch/agent-codaAfter install, skills are available by their track/name path (e.g., agent-coda/app/karpathy-principles).
These skills exist because there is a gap between an AI agent that can write code and one that writes code well. The gap isn't capability — it's craft. Knowing what to do is table stakes. Knowing how to do it with discipline, safety, and awareness of failure modes is what separates deployable work from technical debt.
agent-coda is craft knowledge, not capability knowledge.
Every skill here answers: What does a senior engineer actually know that a capable-but-junior agent doesn't? The answer is usually not a framework or a language feature — it's a pattern, a warning, a checklist earned through watching things break.
| Repo | Focus |
|---|---|
agent-anima |
Inner life skills — identity, memory, trust, growth |
agent-forma |
UI/design craft — components, design systems, UX patterns |
agent-imago |
Generative media — image/video prompting, pipelines |
Creative Commons Attribution 4.0 International (CC BY 4.0)
You are free to use, share, and adapt these skills. Attribution required: link back to this repo.
Maintained by prapanch.