Reference

AI skills library

Each role comes as three files: a Claude Skill for .claude/skills/, a GitHub Copilot custom chat mode for .github/chatmodes/, and a config to paste into ChatGPT's Custom GPT builder. Each file follows its platform's spec and installs as is.

How to use one

Claude Skill: save the file as .claude/skills/<name>/SKILL.md in your project (or ~/.claude/skills/<name>/SKILL.md for every project), restart Claude Code, then call it by name or let it trigger on its own.

Copilot chat mode: save the file as .github/chatmodes/<name>.chatmode.md in your repo. In VS Code's Copilot Chat, pick it from the chat-mode dropdown.

ChatGPT Custom GPT: open Create a GPT in ChatGPT, switch to the Configure tab, and paste the Name, Description, and Instructions fields exactly as given.

Engineering

Backend EngineerAPI design, data modeling, and the failure modes that only show up under load.Cloud/Platform ArchitectBlast radius, cost attribution, and the multi-account tradeoffs that are invisible until an incident.Data EngineerPipelines that stay correct on replay, backfill, and the day a source schema changes without warning.DevOps/SRE EngineerBlast radius, rollback safety, and alerts that page a human only when a human can do something about it.Frontend EngineerAccessibility, state management, and the render-performance bugs that only show up with real data.Full-Stack EngineerThe seam between frontend and backend -- contract drift, auth state in two places, and API calls that don't match what the UI actually needs.ML/AI EngineerThe gap between a model that scores well offline and one that's quietly wrong in production.Mobile EngineerOffline-first sync conflicts, background-execution limits, and the app-store review gotchas that stall a release.QA/Test EngineerTests that catch real regressions, not coverage numbers that go up while confidence goes down.Security EngineerAuth/authz bypass patterns, secrets handling, and threat modeling before the design is locked in.

Product & design

Product Designer / UXDesigns validated by what users said in an interview instead of what they actually did, and accessibility treated as a pass at the end instead of a constraint from the start.Product ManagerFeature requests mistaken for the job underneath them, roadmaps set by the loudest stakeholder instead of the strongest evidence.Technical Program ManagerCross-team dependencies with no real owner, and status reports that say "on track" right up until the week they don't.

Marketing & growth

Brand & Marketing LeadPositioning that could be pasted onto a competitor's site unchanged, and a short-term campaign win bought with brand equity nobody's tracking the cost of.Content Marketing ManagerContent that ranks but never converts, and SEO advice cargo-culted from a blog post instead of how ranking actually works.Growth / Performance Marketing ManagerLast-click attribution taking credit paid retargeting didn't earn, and an A/B test called a winner two days early.

Data & analytics

Data Analyst / BIMetrics that mean the same thing on every dashboard, and correlations worth actually acting on.

Leadership & operations

CEO / FounderTelling a real strategic decision apart from busywork dressed up as strategy, and killing what isn't working before it kills the runway.CTOTechnical debt worth paying down now versus debt that's fine to carry, and architecture decisions that are expensive to reverse treated differently from the ones that aren't.Customer Success / Solutions EngineerThe customer's stated request told apart from their actual underlying problem, and churn signals caught before anyone says the word "cancel."Engineering ManagerA real performance problem told apart from a mismatched-expectations problem, and feedback that's direct without being softened into uselessness.Technical WriterDocs that are accurate but useless because they describe the implementation instead of the task the reader actually came to do -- and the gap between "we wrote docs" and someone being able to follow them.

Why an AI spend tracker built a skills library

These are free, with no signup. Merit AC tracks what happens after a team adopts AI tools: whether the spend on Copilot, Claude or ChatGPT seats produces shipped work or inflates usage numbers with slop. The dashboard rolls spend, outcomes and rework into one score per person and per team.