--- title: Eval-driven development | daily.dev description: Firebase describes how it builds and measures 'agent skills'—structured cheat sheets that help AI coding agents write correct Firebase code alongside the... canonical: https://daily.dev/posts/eval-driven-development-uiieup7bb twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Eval-driven development | daily.dev og:description: Firebase describes how it builds and measures 'agent skills'—structured cheat sheets that help AI coding agents write correct Firebase code alongside the... og:url: https://daily.dev/posts/eval-driven-development-uiieup7bb og:image: https://api.daily.dev/og/posts/UiieuP7Bb.png og:image:alt: Eval-driven development og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Eval-driven development **[Firebase](https://daily.dev/sources/firebase-blog)** · 8 min read · 19 upvotes · 1 comments ## Summary Firebase describes how it builds and measures 'agent skills'—structured cheat sheets that help AI coding agents write correct Firebase code alongside the Firebase CLI and MCP servers. The team shifted to 'eval-driven development': writing automated evals (single-skill, activation, and end-to-end multi-skill tests) before writing skills, establishing baseline failure rates, then iterating on skill content until agents pass consistently. Data shows agents equipped with tuned skills went from a 31.7% to 78.0% pass rate, while using fewer input/output tokens and less time. The same eval suite is also used to find and fix friction in the Firebase CLI itself, such as confusing help text and blocking interactive prompts. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much does adding agent skills improve an AI coding agent's success rate on Firebase tasks? Pass rate improved from 31.7% without skills to 78.0% with tuned Firebase agent skills. Average input tokens dropped from 285.1k to 169.5k, output tokens fell from 9.5k to 5.8k, and average task duration decreased from 144.3 seconds to 101.2 seconds, indicating agents worked more efficiently with less wasted context searching. _Track how AI coding agents perform with and without structured guidance by following developer tooling updates on daily.dev._ ### What are the different types of evals used to test AI agent skills? Three eval types are used: single skill evaluations that test one skill's core functionality against 5-10 specific cases per critical user journey, skill activation evaluations that check whether an agent's frontmatter triggers the skill on relevant prompts without over-activating, and end-to-end multi-skill evaluations that test all skills together against a real deployed project. _Developers designing agent evaluation pipelines can compare approaches like this one on daily.dev._ ### What is the difference between an AI agent skill and official product documentation? Documentation provides comprehensive reference material, while a skill is a highly focused, outcome-driven cheat sheet designed to onboard an agent to a product, enforce core principles, highlight best practices, and give preemptive guidance where agents commonly struggle, such as writing secure Firestore database rules or provisioning cloud resources correctly. _Anyone building agent-friendly tooling can follow this kind of guidance pattern via daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@agustinbarrientos** · 0 upvotes > Useful work, especially the shared view of pass rate, token use, and elapsed time. I'd add human-review disagreement by judge category for teams calibrating the same suite. ## Similar posts on daily.dev - [Agent-driven development in Copilot Applied Science](https://daily.dev/posts/agent-driven-development-in-copilot-applied-science-38qqgozp2) · GitHub Blog · 10 upvotes · 0 comments - [Improving skill-creator: Test, measure, and refine Agent Skills](https://daily.dev/posts/improving-skill-creator-test-measure-and-refine-agent-skills-obmf0r4sa) · Claude · 75 upvotes · 3 comments - [Driving the Agent Quality Flywheel from Your Coding Agent](https://daily.dev/posts/driving-the-agent-quality-flywheel-from-your-coding-agent-gnzecqdjn) · Google Developers · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#firebase](https://daily.dev/tags/firebase) [View this post on daily.dev](https://daily.dev/posts/eval-driven-development-uiieup7bb) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Eval-driven development","url":"https://daily.dev/posts/eval-driven-development-uiieup7bb","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/eval-driven-development-uiieup7bb"},"datePublished":"2026-08-23T12:17:54.485Z","dateModified":"2026-09-14T08:46:22.459Z","description":"Firebase describes how it builds and measures 'agent skills'—structured cheat sheets that help AI coding agents write correct Firebase code alongside the...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3be9090aa7c055fdf081301dcbf772d5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3be9090aa7c055fdf081301dcbf772d5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Firebase","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Firebase","logo":"https://media.daily.dev/image/upload/s--i395oqZH--/f_auto,q_auto/v1787487448/logos/firebase-blog?_a=BAMAMicg0","url":"https://daily.dev/sources/firebase-blog"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/eval-driven-development-uiieup7bb","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":19},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-agents,mcp,firebase","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Firebase","item":"https://daily.dev/sources/firebase-blog"},{"@type":"ListItem","position":3,"name":"Eval-driven development"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/eval-driven-development-uiieup7bb","comment":[{"@type":"Comment","text":"Useful work, especially the shared view of pass rate, token use, and elapsed time. I’d add human-review disagreement by judge category for teams calibrating the same suite.","datePublished":"2026-08-26T06:05:22.534Z","url":"https://daily.dev/posts/UiieuP7Bb#c-6J0tpJnRd","author":{"@type":"Person","name":"Agustin Barrientos","url":"https://daily.dev/agustinbarrientos","image":"https://media.daily.dev/image/upload/s--5ayxQnqn--/f_auto/v1788281802/avatars/avatar_wQYYVe5Tbj0NJ7C7qPoa8?_a=BAMAMicg0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/eval-driven-development-uiieup7bb#faq","mainEntity":[{"@type":"Question","name":"How much does adding agent skills improve an AI coding agent's success rate on Firebase tasks?","acceptedAnswer":{"@type":"Answer","text":"Pass rate improved from 31.7% without skills to 78.0% with tuned Firebase agent skills. Average input tokens dropped from 285.1k to 169.5k, output tokens fell from 9.5k to 5.8k, and average task duration decreased from 144.3 seconds to 101.2 seconds, indicating agents worked more efficiently with less wasted context searching. Track how AI coding agents perform with and without structured guidance by following developer tooling updates on daily.dev."}},{"@type":"Question","name":"What are the different types of evals used to test AI agent skills?","acceptedAnswer":{"@type":"Answer","text":"Three eval types are used: single skill evaluations that test one skill's core functionality against 5-10 specific cases per critical user journey, skill activation evaluations that check whether an agent's frontmatter triggers the skill on relevant prompts without over-activating, and end-to-end multi-skill evaluations that test all skills together against a real deployed project. Developers designing agent evaluation pipelines can compare approaches like this one on daily.dev."}},{"@type":"Question","name":"What is the difference between an AI agent skill and official product documentation?","acceptedAnswer":{"@type":"Answer","text":"Documentation provides comprehensive reference material, while a skill is a highly focused, outcome-driven cheat sheet designed to onboard an agent to a product, enforce core principles, highlight best practices, and give preemptive guidance where agents commonly struggle, such as writing secure Firestore database rules or provisioning cloud resources correctly. Anyone building agent-friendly tooling can follow this kind of guidance pattern via daily.dev."}}]} ```