AI Skills & Templates
Turn your recurring work into reusable AI skills and templates: Claude Agent Skills, Projects, Custom GPTs and Gems, PRD and report templates, a PM skill library, testing skills with evals, and a case study → PRD → presentation skill.
AI Skills & Templates
Turn your recurring work into reusable AI skills and templates: Claude Agent Skills, Projects, Custom GPTs and Gems, PRD and report templates, a PM skill library, testing skills with evals, and a case study → PRD → presentation skill.
In one line
A skill is a reusable method you teach an AI once (instructions, templates, examples, scripts) so it does a recurring task the same good way every time. A tool gives the AI access or actions; memory and evals are parts of the system around it.
The 8 classes
Skills, tools, memory and evals: the system
Know the parts of an AI assistant and where a skill fits.
- Skill = method; tool = access/action; memory = context; evals = checks.
- Package anything recurring.
- Improve the skill, not each output.
Templates: the foundation of every skill
Design templates that make outputs consistent and complete.
- Templates = sections + guidance + examples + checklist.
- Start from proven templates; adapt to your team.
- The checklist becomes your eval.
Where to build skills: Projects, GPTs, Gems and Agent Skills
Choose the right place to package a skill.
- Projects for ongoing context; GPTs/Gems for shareable assistants.
- Agent Skills package instructions + files + scripts, loaded when relevant.
- Write a clear description so the skill triggers at the right time.
Writing great skill instructions
Write instructions that produce consistent, high-quality results.
- Ordered steps, missing-info rules, examples, checklist.
- Keep the core short; reference files for detail.
- The checklist doubles as your eval rubric.
A PM skill library
Build a set of skills that covers your core PM work.
- Cover the PM lifecycle with 8–13 skills.
- Start with the 3 you use most.
- Share team skills; keep personal ones private.
Connecting skills to sources and tools
Ground skills in your documents and let them act through tools.
- Ground skills in real sources.
- Use scripts for exact calculations.
- Approval before external actions.
Testing skills with evals
Prove a skill works and stays working.
- Every skill needs an eval set.
- Mix exact checks and rubric checks.
- Re-run on every change.
Project: case study → PRD → presentation skill
Build a portfolio-ready reusable skill end to end (e.g. for a VitaFit-style assignment).
- Skills + sources + evals = reliable AI workflow.
- Test on several case studies, not one.
- Show before/after eval results in interviews.
Interview questions
What is an AI skill?
A reusable method (instructions, templates, examples, scripts) that makes an AI perform a recurring task consistently.
Skill vs tool?
A skill is how to do something; a tool gives access or performs an action, such as search, database query or ticket creation.
How would you build a PRD skill?
Template + ordered steps + missing-info rule + good example + checklist, connected to source docs, tested with an eval set, iterated on failures.
Projects vs Custom GPTs vs Agent Skills?
Projects hold context for ongoing work; GPTs/Gems are shareable assistants; Agent Skills are modular methods loaded when relevant across surfaces.
How do you test a skill?
Run it on 10–30 realistic inputs with exact checks (names, sections, numbers) and rubric checks (clarity, evidence), after every change.
Which PM tasks would you package first?
The most frequent and template-driven: PRDs, weekly updates, interview synthesis and teardowns.
How do you avoid wrong numbers in AI documents?
Require citations to the source, use helper scripts for calculations, and run faithfulness checks.
What makes skill instructions good?
Ordered steps with outputs, rules for missing info, examples, a checklist and short core text with reference files.
