strategysoul

ai-concept-explainer

Explain an AI or LLM concept at the depth the person actually needs — working definition, the mechanism, where it breaks, and the decision it changes. Use when the user asks what something means (embeddings, RAG, temperature, context window, fine-tuning, agents, evals, tokens, hallucination), says they don't understand an AI concept, or asks how something works under the hood.

ai-practice-project

Design a small hands-on project that teaches an AI or LLM concept by building it — scoped to the hours available, with a milestone ladder, what to observe at each stage, and cost guardrails. Use when the user wants to practice or apply an AI concept, asks for a project idea or exercise to learn LLMs, RAG, agents, prompting, or evals, or says they understand something in theory but have not built it.

ai-research-digest

Cut an AI paper, model release, or announcement down to what actually changed and whether it affects you — the claim, the evidence behind it, what is genuinely new, and what to do now versus watch versus ignore. Use when the user shares an AI paper, release note, changelog, or announcement, asks whether something matters or is hype, or wants help keeping up with AI news.

ai-study-plan

Build a structured plan for learning an AI topic — diagnose the real starting level, set a capability target, sequence units by dependency, and attach an artifact to each one. Use when the user asks how to learn AI, LLMs, prompting, RAG, agents, evals, or machine learning, wants a study plan or learning path or roadmap, or asks where to start with an AI topic.

ai-tool-evaluation

Decide whether a new AI tool, model, or technique is worth adopting — define the job to test it on, set decision criteria before trying it, run an honest comparison against what you already use, and record the decision with a revisit trigger. Use when the user asks whether to adopt or switch to an AI tool or model, wants to compare options, asks if something is worth trying, or is evaluating a vendor or framework.

Data & Documents Listed

bug-report

Turn a vague complaint or observed defect into a reproducible bug report — steps to reproduce, expected vs actual, environment, evidence, and a severity/priority call. Use when the user says something is broken, wants to file a bug, needs to write up an issue or defect, or asks how to triage or reproduce a reported problem.

deep-research

Answer a decision-grade question by researching it across several angles at once, attacking every finding before believing it, and producing a sourced report ranked by confidence — every claim carrying a link and a date. Use when the user asks to research or investigate a question, wants market or competitive or user research, asks what the landscape looks like, asks you to find out whether something is true, or needs evidence behind a product decision rather than an opinion.

Testing & QA Listed

e2e-testing

Execute a test pass against a running build — a local server, a preview or PR deployment, or a live dev environment — by driving a real browser — walk each scenario as written, capture screenshots and console and network evidence, and report pass, fail, blocked, or flaky with a bug report for every failure. Use when the user asks to actually run the test scenarios, test the app in a browser, click through the flows, do a QA pass on a build, verify a fix end to end, or check that something works in the running app rather than on paper.

prd-drafting

Draft a Product Requirements Document from a feature idea, problem statement, or rough notes — problem, users, goals, scope, a step-by-step flow walkthrough covering both what the user sees and how data moves through the system, requirements, success metrics, and open questions. Use when the user asks to write a PRD, spec a feature, document requirements, turn an idea into a spec, or review an existing PRD for gaps.

Web & Frontend Listed

prototype-creation

Design and build a prototype that tests a product idea before it is built — takes a PRD, story, or rough idea as its source, picks the right fidelity, states the question it must answer, and produces a working single-file clickable HTML prototype or a wireframe spec. Use when the user asks to prototype a PRD or feature, wants a mockup, clickable demo, wireframe, or proof of concept, wants to validate an idea cheaply before development, or wants a prototype revised to fix defects found when it was verified against the stories and scenarios.

Testing & QA Listed

prototype-verification

Check a prototype against the user stories and test scenarios it is supposed to satisfy — build a coverage map, walk each scenario as written, and report every failure sorted into prototype defect, spec gap, or bad scenario. Use when the user asks whether a prototype covers the stories or acceptance criteria, wants a prototype tested or verified against test scenarios or a QA pass, asks which stories have no screen, or wants to know if a spec and a prototype actually agree.

Testing & QA Listed

testing-scenarios

Generate a test pass from a PRD, user story, or acceptance criteria — happy paths, boundaries, error handling, permissions, and state-transition cases, each with steps and expected results, traced back to the source spec. Use when the user asks to test a PRD or feature, wants test cases, test scenarios, a QA checklist, edge cases, or asks how to verify something before release.

user-story-creation

Break a PRD, feature, or epic into user stories with acceptance criteria — vertically sliced, INVEST-checked, with Given/When/Then criteria, sizes, and traceability back to the source spec. Use when the user asks to turn a PRD into stories, wants user stories, backlog items, tickets, acceptance criteria, help splitting an epic, or help making requirements sprint-ready.

skill-name-in-kebab-case

One or two sentences. Say what the skill produces, then when to use it — include the phrases a user would actually type, because this description is the only thing the model sees when deciding whether to load the skill.