The AI Starter Skill Pack is five small Agent Skills that help you start, get unstuck, learn by doing, build something real, and make honest progress. No hype. No fifty-tool shopping list. No promise that prompt tricks land you a job.
npm install github:Jeff-Kazzee/ai-starter-skill-packWorks with Claude Code, Codex, and other Agent Skills runtimes.
They need help answering five ordinary questions — and a sensible next action for each. Each skill owns one question.
Use them in order, or reach for whichever one fits. They deliberately overlap at the handoffs — onboarding picks a first step, the tutor teaches, the project skill scopes a build, the career skill weighs the evidence, and the debugger repairs one model interaction.
ai-beginner-onboardingI want to learn AI, but I only have a phone and ten minutes on low-energy days.
Read the skillprompt-debuggerClaude keeps ignoring the question after a long report. Diagnose the prompt.
Read the skillpersonal-ai-tutorTeach me to use coding agents without letting them write code I cannot explain.
Read the skillai-project-ideaGive me a weird but useful AI project I can finish this weekend with public data.
Read the skillno-bs-ai-careerWhat can I honestly put on my resume after building three personal AI demos?
Read the skillSmall, but not slight — every skill ships with reference methods, reusable templates, and trigger-eval tests.
Each skill is a small operating procedure. That's why the same skill behaves well across very different questions — and knows when to step aside.
A description tuned so the right skill activates from a natural question — and only then.
Clear routing to the other skills, so onboarding does not try to do the tutor’s job.
Step-by-step instructions, not a clever wording trick that only works once.
Privacy, pacing, and accessibility baked in — repeated inside each skill so it is safe alone.
A consistent, inspectable result instead of a wall of improvised text.
Deeper methods and reusable worksheets the skill can pull in when needed.
Positive, negative, implicit, and boundary cases so a skill fires at the right time.
Instruction-only. No scripts, no paid API required, MIT-licensed, yours to fork.
Unless you say otherwise, every skill follows the same calm defaults — so the first step is small, private, and actually doable.
One assistant you already have. No paid APIs, agents, or vector databases as a first step.
20–30 minutes a day, with a 10-minute low-energy option for hard days.
Sample or redacted data first, before anything personal, client, school, or employer related.
Model output kept separate from verified fact — assumptions are always labelled.
One small finished artifact over a giant unfinished curriculum.
Accessibility, fatigue, and budget treated as design constraints — not character flaws.
A skill is one folder with a SKILL.md at its root. Copy the ones you want, or install the whole pack from GitHub.
Exposes every skill under node_modules/ai-starter-skill-pack/skills/.
npm install github:Jeff-Kazzee/ai-starter-skill-packOr copy any folder from skills/ straight into your runtime's skills directory — each skill works on its own.
Drop a skill folder into your project or personal skills directory.
.claude/skills/ · ~/.claude/skills/Place skills in the repository or user skills directory.
.agents/skills/ · ~/.agents/skills/Upload versioned skill bundles for use with supported tool runtimes.
Import via the Skills interface — it supports the standard Skill format.
No generic folder loader. Paste a SKILL.md into project instructions as an approximation.
Discovery paths are runtime-specific and change. Check the current docs for your tool before deploying — the links are in the footer.
Native runtimes pick the right skill from your question. Here's the kind of thing people actually say.
I want to learn AI, but I only have a phone and ten minutes on low-energy days.
Claude keeps ignoring the question after a long report. Diagnose the prompt.
Teach me to use coding agents without letting them write code I cannot explain.
Give me a weird but useful AI project I can finish this weekend with public data.
What can I honestly put on my resume after building three personal AI demos?
Fewer people wandering through tool lists. More people completing one safe workflow, one real learning loop, one inspectable project, and one honest next step.