utopia-product

Plugin utopia-product from marketplace skills in The-Utopia-Studio/skills. Commands read from the repo's manifest — 4 days ago.

Skills (134)

• ab-test-analysis

  Analyzes A/B and split test results: conversion rates, lift, p-value, confidence interval, sample size, sample ratio mismatch and a ship / extend / stop call. Use when the user pastes control and variant numbers, asks if a test is significant, whether the lift is real, or whether to ship the variant

• agent-concierge-probe

  Runbook for having an AI agent run a real, mapped task end to end on several cases, producing cost per outcome priced to the cent and a frontier map of steps it does unaided vs needs a human. Use when the user wants to know what one outcome costs, whether a service can be automated, or what the agen

• agent-design

  Produces a four-part spec for one production agent: role (one owned decision), tools, memory layer (CLAUDE.md, skills, lessons.md, trace archive) and eval. Use when a fellow says design an agent for X, asks what tools and memory an agent needs, or says the agent never remembers corrections. Not for

• agents-md-writer

  Writes AGENTS.md, CLAUDE.md-style context and .cursorrules or .cursor/rules so coding agents (Claude Code, Cursor, Codex, Copilot) understand a repo's stack, commands, structure, conventions and boundaries. Use when the user asks to write or improve AGENTS.md, CLAUDE.md or .cursorrules, make a repo

• ai-solution-proposal-canvas

  Builds a one-page proposal canvas for an AI product or feature: business and customer outcomes, problem, solution hypothesis, positioning, assumptions, PESTEL risks, value case and success metrics. Use when the user is proposing an AI feature to executives, pitching for budget, or deciding whether a

• analytics-tracking

  Writes and audits analytics tracking plans: event names and properties, naming conventions, UTM parameters, conversion and funnel events, and GA4, GTM, Mixpanel, Amplitude, PostHog or Segment setup. Use when the user asks for a tracking plan, event tracking for a launch, conversion or attribution tr

• asd-ste100

  Writes, rewrites or audits technical documentation in ASD-STE100 Simplified Technical English (Issue 9). Use when the user asks for STE or Simplified Technical English, mentions ASD-STE100, S1000D or ATA iSpec 2200, or wants maintenance manuals, work instructions, SOPs, safety warnings or procedures

• bottoms-up-market-sizing

  Builds a market size from countable units up: atom x reachable population x frequency x value per event x capture fraction, every rung sourced and evidence-weighted, with top-down TAM only as a final sanity check. Use when the user says don't give me a TAM, size this bottom-up, show the arithmetic,

• brand-graphics

  Designs brand graphics: logos and AI logo generation, corporate identity mockups (business cards, letterheads, signage), social and ad banners at exact platform sizes, HTML slide decks with Chart.js, and SVG icon sets. Use when the user asks for a logo, brand identity or CIP mockups, a banner or cov

• brand-guidelines

  Creates, updates and reviews brand guidelines: brand voice and tone, visual identity, messaging frameworks, asset naming and validation, color extraction, and syncing brand-guidelines.md into design tokens. Use when the user asks for a style guide, tone of voice, brand consistency review, on-brand m

• build-vs-buy-boundary

  Decides build vs buy for each component of an AI product: what the studio platform (Daedalus) provides and the fellow inherits, what the fellow must build as their moat, and what to rent, in a boundary map. Use when the user asks whether to build their own eval harness, RAG pipeline, model router, g

• business-model

  Builds a business model canvas for a product or venture: Lean Canvas, Startup Canvas or the 9-block Business Model Canvas, ending with the riskiest assumptions and cheap validation experiments. Use when the user asks for a Lean Canvas, startup canvas, BMC, business model canvas, or how a new venture

• cli-agent-dx-score

  Scores a CLI 0-21 on seven agent-DX axes (machine-readable output, JSON input, schema introspection, context-window discipline, input hardening, safety rails, agent knowledge packaging) with fixes. Use when the user asks how agent-friendly or AI-ready their CLI is, wants it rated for AI agents, or s

• cloud-cost-optimizer

  Cuts cloud and hosting bills: finds cost traps (always-on EC2, multi-AZ RDS, NAT gateways, log retention, S3 lifecycle, Vercel and Railway tiers), idle or oversized resources, quick wins with monthly savings, and billing alerts. Use when the user has a high or rising AWS, GCP or Vercel bill, asks wh

• compound-system-architecture

  Designs the build architecture of an AI product as a compound system: input, router, retrieve/reason/act, validation and guardrails, output over a data layer, with a Data 40 / Model 20 / Orchestration 30 / Interface 10 effort split and model-swap points. Use when asked how to architect an AI product

• concept-council

  Adversarial teardown of a raw one-line product idea by five lenses (Skeptic, Buyer, Incumbent, Engineer, Regulator), returning a Gap Map with the one load-bearing gap and a first probe that costs under a week. Use when someone pitches a new idea and says tear this apart, rip it apart, what am I miss

• concept-four-lenses-test

  Scores one shaped product concept on four build risks (desirability, usability, feasibility, viability), each graded only by its own evidence tool, with a go/no-go that fails if any lens fails. Use when the user asks if a concept is desirable, usable, feasible or viable, wants a go or no-go on a sol

• concierge-probe

  Runbook for a concierge MVP: the founder delivers the outcome by hand to one real customer at a price set in advance, producing a concierge log and a willingness-to-pay score on the evidence ladder. Use when the user says do it by hand first, concierge test, deliver it manually, or wants to test if

• continuous-discovery-engine

  Builds and maintains a living opportunity solution tree (outcome, opportunities, solutions, tests) fed by customer interviews, production traces and eval telemetry, each with a cadence and a named owner. Use when the user says set up discovery, continuous discovery, opportunity solution tree, keep l

• current-state-map

  Maps a workflow exactly as it is done today and prices every step: touch time, handoffs, delay, failure and rework, totalled into cost per unit with sources and the most expensive step. Use when the user lists an as-is process with times or rates and asks how it is done today, what one unit really c

• data-rights-clause

  Drafts the data-rights clause for a first paid pilot: tests whether the data is worth a clause, then writes language on data covered, retention, derived data, training rights and survival, with GCC/MENA points flagged for counsel. Use when the user asks who owns the data, what to sign at the pilot,

• dataset-builder

  Turns real field artefacts (scanned documents, logs, tickets, emails, expert corrections) into a labelled dataset and 20+ golden eval cases, with ground-truth labels and four gates: provenance, ground truth, unreachable by the model, held out. Use when the user has pilot data or corrections and asks

• deploy-to-vercel

  Deploys apps and websites to Vercel and manages them: link the project, ship preview or production deploys, use a VERCEL_TOKEN when login is impossible, set env vars, inspect builds and logs, add domains. Use when the user says deploy my app, deploy to Vercel, push this live, give me a preview link,

• deployment-engineer

  Sets up CI/CD and deployment infrastructure: GitHub Actions for lint, tests and deploys, a production Dockerfile and docker-compose, separate dev/staging/production environments with env vars, and health checks. Use when the user asks to set up CI/CD or GitHub Actions, dockerize an app, create stagi

• design-of-experiments

  Designs a multi-factor experiment that learns the most from the fewest runs or cohorts: full and fractional factorial, screening and response-surface designs, with design matrix, randomisation, blocking, power and an analysis plan. Use when testing several changes at once ("5 onboarding changes, onl

• design-tokens

  Builds a design-token architecture: primitive, semantic and component token layers, token JSON, generated CSS variables, spacing and type scales, component state specs, and checks for hard- coded values. Use when the user asks for design tokens, a token system, CSS variables from brand colors, a com

• dev-tool-integrations

  Connects a startup's dev tools: GitHub and Vercel notifications into Slack, Linear issues linked to branches and PRs, Sentry alerts into Slack, with step-by-step setup and a connectivity map. Use when the user asks to connect GitHub, Slack, Linear, Jira, Sentry or Vercel, set up PR, deploy or error

• diagram-design

  Creates technical and product diagrams as standalone HTML with inline SVG: architecture, cloud and network topology, flowchart, sequence, state machine, ER, timeline, swimlane, quadrant, tree, layer stack, venn and pyramid, in light, dark or sketchy styles. Use when the user asks for an architecture

• discovery-interview-prep

  Plans and scripts customer discovery interviews: research goal, segment, method, and a ready Mom Test script with JTBD questions, probes, willingness-to-pay and a note-taking template. Use when the user asks for an interview script or guide, to prepare or plan customer or user interviews, or mention

• discovery-process

  Runs one full product discovery cycle from problem hypothesis to validated solution: frame the problem, research, synthesize, generate and test solutions, then decide build, pivot or kill. Use when the user wants a structured discovery plan, needs to investigate a drop in activation or retention, or

• efecto-graphic-design

  Designs graphic assets in the Efecto design tool via its MCP server: decks, posters, email headers, OG cards, business cards, resumes, menus, infographics, invitations, newsletters. Use only when the user mentions Efecto or Efecto MCP tools (create_session, add_section, get_document) are available.

• efecto-social-media

  Designs social media graphics in the Efecto design tool: Instagram posts and carousels, stories, YouTube thumbnails, TikTok covers, LinkedIn carousels and Pinterest pins, with platform sizes and safe zones. Use when asked to design an Instagram carousel, make a thumbnail, or design a social post gra

• efecto-web-design

  Designs web pages and app UIs live in the Efecto design tool via its MCP server: sessions, JSX and Tailwind layouts, artboards and real-time changes the user watches in the browser. Use only when the user mentions Efecto or Efecto MCP tools (create_session, add_section, get_document) are available.

• eval-first-spec

  Turns a validated wedge into a scoreable build spec: a one-sentence job, 20 pass/fail golden cases from real artefacts, an L0-L4 autonomy level with failure taxonomy and acceptable failure rate per mode, and a cost-per-outcome budget. Use when a fellow says spec the build, scope the v1, define scope

• evidence-ladder

  Scores how strong validation evidence really is by what people did, not said: places each signal on one ladder (money 1.0, behaviour 0.7, artefact 0.5, verbal 0.3, opinion 0.1) and returns a weighted table and the cheapest probe to raise each one. Use when the user lists quotes, sign- ups, LOIs, pil

• explicit-vs-tacit-capture

  Splits a body of expertise (SOPs, manuals, rubrics, expert transcripts) into explicit procedures a stranger could follow and tacit judgment that only an expert can show, and names the tacit half as the defensible product. Use when the user asks to codify expertise, what is teachable vs judgment, or

• falsify-plan-assumptions

  Falsifies the few load-bearing beliefs under a chosen plan or wedge: restates each as a null hypothesis, designs the cheapest observation that would disprove it, sets a kill signal in advance, and ranks beliefs by chance of being wrong times impact. Use when the user asks what has to be true, how to

• fellow-level-ladder

  Assesses which Icarus level one fellow holds (Literate, Practitioner, Operator, Frontier, Author) by checking which required exit artefacts actually exist, never by self-assessment, and names the next artefact and the skill that produces it. Use when the user asks how a fellow is doing, what level t

• fellow-path-router

  Routes a new fellow through the Icarus stages by situation: classifies them as Type A (blank page), B (traction, no moat) or C (mature product) on evidence, then returns a keep / trim / skip stage plan and a think:build:test ratio. Use when a fellow asks where to start, which stages apply or can be

• first-mocks

  Turns a product concept into a clickable no-code prototype in one afternoon: Crazy 8s, paper or Miro flow, digital mock, then a clickable hybrid in v0 or Figma Make, with the one reaction the mock must provoke. Use when the user says mock it up, run Crazy 8s, make a clickable prototype, or turn an i

• greptile-review-loop

  Loops a pull request through Greptile until it scores 5/5 confidence with zero unresolved comments: trigger a review, fix every actionable comment, push, re-trigger, repeat, on GitHub or GitLab. Use when the user mentions Greptile, a Greptile confidence score, or wants Greptile's review comments fix

• guardrail-design

  Designs the guardrails that stop an AI product doing the wrong thing: places each failure mode on a cost-of-error by volume matrix, sizes a three-layer stack (code rules, confidence threshold, human sign-off) and sets the auto-approve threshold and sign-off triggers. Use when asked to design guardra

• hf-cli

  Runs Hugging Face Hub operations with the hf CLI (replaces huggingface-cli): login, cache, download and upload of models, datasets and Spaces, repos, buckets, PRs, collections, webhooks, HF Jobs and Inference Endpoints. Use when the user mentions hf or huggingface-cli or wants a Hub operation from t

• huggingface-community-evals

  Runs benchmark evaluations of Hugging Face Hub models on local GPUs with lighteval or inspect-ai (GSM8K, HellaSwag, MMLU and similar) and picks the backend (vLLM, Transformers or accelerate) with exact commands. Use when the user wants to evaluate or benchmark an HF model locally, mentions lighteval

• huggingface-datasets

  Queries Hugging Face datasets through the Dataset Viewer API: list subsets and splits, preview and paginate rows, search text, filter rows, get parquet links, sizes and column statistics, and run SQL on parquet. Use when the user wants to explore, sample, search or filter a Hub dataset, read its row

• huggingface-gradio

  Builds Gradio web apps and demos in Python: Blocks layouts, tabs, file uploads, streaming chatbots, components and event listeners, and calling Gradio apps on Spaces. Use when the user mentions Gradio or wants a quick Python web UI or demo for a model, PDF, chatbot or ML pipeline. Not for React fron

• huggingface-llm-trainer

  Trains or fine-tunes language and vision-language models with TRL or Unsloth on Hugging Face Jobs cloud GPUs: SFT, DPO, GRPO and reward modelling, dataset checks, hardware and cost estimates, Trackio monitoring and GGUF conversion. Use when the user wants to fine-tune an LLM, run SFT, DPO or GRPO, t

• huggingface-papers

  Looks up a research paper on Hugging Face Papers by arXiv ID or URL: reads it as markdown, summarises contributions and results, and lists the linked models, datasets, Spaces, GitHub repo and project page. Use when the user gives an arXiv ID, arXiv URL or huggingface.co/papers link and asks to summa

• huggingface-tool-builder

  Writes reusable Python, shell or TypeScript scripts that fetch, chain, enrich and export Hugging Face Hub API data: top or trending models by task, licenses, base models, linked papers, model- card metadata, to CSV or JSON. Use when the user wants a script that pulls models, datasets or Spaces data

• huggingface-trackio

  Tracks ML training experiments with Trackio: logs metrics from Python or TRL, fires alerts on loss spikes, NaN gradients or stalls, syncs dashboards to a Hugging Face Space, and retrieves metrics and alerts from the CLI as JSON. Use when the user mentions Trackio, wants to log or monitor training me

• huggingface-vision-trainer

  Trains or fine-tunes vision models on Hugging Face Jobs cloud GPUs: object detection (D-FINE, RT-DETR v2, DETR, YOLOS), image classification (timm, ViT, MobileNet, ResNet) and SAM/SAM2 segmentation, with COCO data prep, augmentation and mAP evaluation. Use when the user wants to train a detector, cl

• identify-risky-assumptions

  Lists and rates the risky assumptions behind a product idea or feature: 8 risk categories for a new venture, or the 4 core risks for a feature, each with confidence and a test. Use when the user asks for the riskiest assumptions, an assumption map or risk assessment, devil's advocate on a feature, o

• impeccable

  Builds, redesigns or studies frontend interfaces so they look designed, not AI-generated, and holds the shared design context the ui-* skills load first. Use when the user asks for a landing page, dashboard, app UI or component with a distinctive look, to upgrade a site to premium or agency-grade qu

• industrial-brutalist-ui

  Builds web interfaces in an industrial brutalist style: Swiss print grids or dark CRT terminal telemetry, extreme type-scale contrast, monospace data, utilitarian color, halftone and scanline effects. Use when the user asks for a brutalist, industrial, terminal, military HUD or blueprint look, or a

• infra-scaling-review

  Reviews whether infrastructure will survive growth: database connection pooling, indexes, migrations and backups, caching and CDN, job queues, statelessness and horizontal scaling, with detection commands and fixes ranked by impact. Use when the user asks if their stack can go from 1k to 50k users,

• ink-terminal-ui

  Renders interactive terminal UIs from JSON specs with @json-render/ink, the Ink renderer for json-render: catalogs, standard and custom components, data binding, visibility and actions. Use when the user works with @json-render/ink, wants to build a terminal UI from JSON, create a terminal component

• invent-product-concept

  Invents an original, non-obvious product concept instead of the generic one every competitor would prompt for: peels the idea to its core need, diverges, anchors each concept in proprietary data, then runs a generic-prompt test. Use when the user's space is crowded with the same idea, or asks what t

• job-in-primitives

  Strips a workflow to the primitive job underneath: deletes every tool, vendor and job-title name, then states the job as information moved, decisions made and liability transferred, with an invariance test. Use when the user describes a process full of tool and team names and asks what job is really

• jobs-to-be-done

  Maps customer jobs to be done (functional, social, emotional), pains and gains in a JTBD / Value Proposition Canvas format from interview notes or research. Use when the user asks for jobs to be done, a JTBD analysis, customer jobs, pains and gains, or why customers hire or fire a product. Not for f

• lean-ux-canvas

  Runs Jeff Gothelf's Lean UX Canvas (v2) for a feature or initiative: business problem, business and user outcomes, users, solutions, hypotheses, what to learn first and the least work to learn it. Use when the user asks for a Lean UX Canvas session, or wants to frame a business problem, surface assu

• metrics-dashboard

  Designs a product metrics dashboard spec: North Star, input, health and business metrics, each with an exact definition, data source, chart type, target and alert threshold. Use when the user asks to build or define a metrics dashboard, set KPIs and alerts, or plan product data monitoring. Not for c

• metrics-that-matter

  Picks and validates the few metrics that matter for a product, especially a launched AI product: one North Star with 3-5 input metrics, an AARRR skeleton, a cohort retention test and cost per outcome. Use when the user asks what metrics to track, to choose a North Star, whether a growing headline nu

• minimalist-ui

  Builds pages in a clean editorial, Notion-like minimalist style: warm off-white canvas, serif headlines with sans body, flat bento grids with 1px borders, muted pastel tags, near-zero shadows and no gradients. Use when the user asks for a minimalist, editorial, Notion-style or document-like UI or a

• moat-design-canvas

  Turns a validated wedge into a defensible moat as four dated gates (wedge, usage, lock-in, data advantage), with an incumbent-won't-copy sentence and monthly falsifiers, in a filled moat canvas. Use when the user asks what the moat is, how the product is defensible, whether it will compound, or whet

• monitoring-setup

  Sets up production monitoring: Sentry error tracking, PostHog product analytics, health checks and alerting, with verification commands and output templates. Use when the user asks to set up monitoring, add error tracking, configure Sentry or PostHog, add analytics or set up alerts. Not for wiring G

• paper-sketch-probe

  Runbook for a 30-minute paper prototype test: draw the workflow as 5-9 panels, hand it to one real operator in silence, log pauses, and report whether the flow was legible and which panel holds the real decision. Use when the user says paper test, sketch it, paper prototype, or wants to check an ope

• pilot-six-term-sheet

  Structures a paid pilot as a six-term sheet (scope, price, data-rights clause, success metrics, kill criteria, conversion terms priced now) and refuses to call it a pilot unless all six are filled. Use when the user asks for pilot terms, a paid pilot, a pilot term sheet, to structure the deal, or to

• post-launch-refine-loop

  Sets up the post-launch improvement loop for a shipped AI product: ship behind human sign-off, track how much customers edit outputs, re-run the full eval set on every new model, and raise autonomy only on eval scores. Use when the user asks what to do after launch, whether to upgrade to a new model

• pr-test-evidence-video

  Records annotated screen-recording proof that a UI fix or feature works (setup, test steps, pass/fail assertions) and posts the video and a results summary to the pull request and the Linear or Jira ticket. Use when the user asks to record proof or a video of a UI flow, show a bug fix works on a PR,

• prd-development

  Writes a product PRD at the right weight: a one-pager, a lean 1-2 page spec, or a full engineering-ready PRD with problem, evidence, personas, solution, metrics, stories, scope and risks. Use when the user asks to write or review a PRD, one-pager, product spec, feature proposal or product brief, or

• pricing-strategy

  Develops pricing and monetization strategy: compares revenue models (freemium, subscription, usage, seat, marketplace), designs tiers, value metric and price points, researches willingness to pay, and prices AI products. Use when the user asks how to monetize, what to charge, about pricing tiers, fr

• prioritization

  Ranks features, initiatives or customer problems with RICE, ICE, weighted scoring or Opportunity Score tied to strategic pillars, giving a ranked list, sequence and cut line; also explains which framework to use. Use when the user says prioritize my backlog, RICE score, rank these initiatives, what

• probe-matrix

  Picks the cheapest honest experiment for one product question: writes it as an XYZ hypothesis, routes it to a paper sketch, Wizard-of-Oz, concierge, agent-concierge or fake-door test, states what that probe cannot tell you, and sets pass/fail in advance. Use when the user asks how to test or validat

• problem-framing-canvas

  Runs the MITRE Problem Framing Canvas: look inward at assumptions and biases, look outward at who has the problem and who is left out, then reframe into a problem statement and How Might We question. Use when the user has a messy problem, wants to reframe a stakeholder request, or needs a clearer pr

• problem-quality-scorecard

  Go/no-go gate on a product problem before any solution: scores it 1-5 on eight evidence-backed dimensions (frequency, budgeted pain, severity, data exhaust, persistence, buyer clarity, wedge, founder edge), out of 40, with pass, redesign or kill. Use when the user asks if a problem is worth building

• problem-statement

  Writes a user-centred problem statement: I am, trying to, but, because, which makes me feel, plus context and a one-line final statement. Use when the user needs to frame a problem for discovery, a PRD or prioritization, align stakeholders on the problem, or turn a feature request back into a user p

• product-as-decision

  Restates a product idea as one decision a named person makes, then quantifies how it gets compressed before to after on one axis of time, effort or autonomy (e.g. six minutes to thirty seconds). Use when the user asks what the product really is, what decision we are changing, to state this as a deci

• product-strategy-stack

  Turns a validated problem into the product strategy stack: vision, strategy, product vision, North Star, OKRs and a dual-track roadmap, each layer resting on the evidence below it. Use when the user has a validated problem and asks to frame the business, go from vision to roadmap, set strategy, or d

• proprietary-data-sourcing

  Finds the proprietary data a product can own that competitors and foundation models cannot reach: sorts candidate sources (logs, work orders, chat threads, sensor feeds, manuals) by whether a model can already get them and keeps workflow-trapped data as the moat seed. Use when the user asks what dat

• proto-persona

  Creates a proto-persona: an assumption-based profile of the target user built from available research, market signals and team knowledge, with goals, pains, behaviours and the gaps to validate. Use when the user needs a working persona or customer profile early, before deep research, or to align a t

• railway
• repo-readiness-scan
• repo-structurer
• roadmap-planning
• saas-economics-efficiency-metrics
• saas-revenue-growth-metrics
• security-auditor
• service-layer-refactor
• shadcn-tailwind-ui
• so-what-stress-test
• stitch-design-taste
• summarize-interview
• synthetic-users
• tacit-knowledge-interview
• tdd-red-green-refactor
• trace-to-interview
• transformers-js
• typed-service-contracts
• ui-advanced-effects
• ui-animation-storyboard
• ui-animation
• ui-audit
• ui-bolder
• ui-color
• ui-critique
• ui-delight
• ui-design-brief
• ui-layout-spacing
• ui-microcopy
• ui-performance
• ui-polish
• ui-quieter
• ui-responsive
• ui-simplify
• ui-typography
• ui-ux-pro-max
• ui-walkthrough
• unit-cost-floor-gap
• unserved-needs-finder
• usability-test-protocol
• user-buyer-decider-map
• user-story-mapping
• user-story-splitting
• user-story
• v1-launch-bar
• value-based-pricing
• vercel-composition-patterns
• vercel-react-best-practices
• vercel-react-native-skills
• walkthrough-video
• wedge-five-questions
• wizard-of-oz-probe
• workflow-design
• working-backwards-press-release

Similar plugins