GadDev

ai-pulse

Produces hype-filtered, technically substantive AI news digests and research-paper roundups instead of relaying a wall of product headlines uncritically. Defaults to engineering depth — architecture changes, training methods, infra/serving efficiency, benchmark internals — not just pricing, availability, and marketing copy. Use this whenever the user asks for an AI news roundup or digest — e.g. "what's new in AI," "catch me up on AI news," "any AI news today," "give me an AI update," "anything technical happening in AI" — or asks about recent AI research or papers — e.g. "any interesting AI papers lately," "what's new in research," "latest papers on [topic]." For a question about one specific named model/tool/release, use `ai-release-triage` instead — this skill is for broad, multi-item roundups.

ai-release-triage

Runs a structured evaluation of a new AI model, tool, agent framework, or coding-agent technique instead of repeating launch-blog hype. Use this skill whenever the user mentions a newly announced or released AI model/tool/framework, asks whether it's "actually good," "worth switching to," "hype or real," or how it compares to what they're using now, or wants help deciding whether to adopt something in their agent/coding stack. Trigger even without the words "evaluate" or "triage" — e.g. "is this new model actually better," "should I switch my agent to X," "what's real about this release."

brainstorm-to-blueprint

Turns a raw feature brainstorm (usually a messy dump pasted from ChatGPT or notes) into three grounded planning artifacts — a constitution, a roadmap, and a techstack — without inventing scope, decisions, or technology the user never stated. Use this whenever the user pastes brainstorm/ideation content and asks to structure, formalize, or "clean up" it into a spec, constitution, roadmap, or tech-stack doc — e.g. "turn this brainstorm into a spec," "make a constitution/roadmap/techstack from this," "structure my idea dump," "format this into planning docs." Compatible with Spec Kit (.specify/) and OpenSpec (openspec/) if the project uses them. For evaluating an AI product/release instead, use ai-release-triage.

Code & Development Listed

evidence-tiering

Shared credibility framework for weighing any AI-related claim — a benchmark result, a model/tool announcement, a research paper, a demo, or a viral thread — by the quality of its source rather than how confidently it's stated. Use this any time you're about to repeat a claim about AI capabilities, performance, or news without having checked where it actually came from. Other skills (ai-release-triage, ai-pulse) call into this for their source-vetting step, but use it standalone too whenever a claim about AI needs a credibility check before you pass it on.

Web & Frontend Listed

design-skill

Guides the user through designing and authoring a new Claude Code Skill by asking targeted questions and formatting their answers into a working SKILL.md — covering trigger design (the description field), scope (one workflow per skill), structure, worked examples, and bundled reference files. Use whenever the user wants to create, author, design, draft, or scaffold a new skill — e.g. "help me build a skill for X," "design a skill," "how do I write a SKILL.md," "turn this workflow into a skill," "why isn't my skill triggering." Not for editing an unrelated existing skill's content (just edit it directly) — this is for the design/authoring process itself, including diagnosing why a skill isn't loading.

prompt-architect

Iterative guardrail and assistant for writing effective prompts that produce consistent, high-quality AI outputs. Use when the user says things like "my prompt isn't working," "help me write a better prompt," "improve this prompt," "make this clearer," "this AI keeps giving me the wrong output," "how do I stop this from hallucinating," or pastes a raw prompt and wants feedback. Runs an analyse → improve → score loop until the prompt passes a readiness check, applies structured techniques — system prompt design, XML tags, few-shot examples, output constraints, and hallucination-grounding techniques — and flags when an output needs human review regardless of prompt quality. Do NOT use for single-file code edits, model evaluation questions, or AI news — those go to other skills.