Skip to content
maxgfrPublic

About

Chrome extension for web developers: cache & site data, page inspection (JSON-LD, headers, Web Vitals…), request rules, utilities and AI analysis with your own model

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

devkit

A Chrome extension for web developers. Everything in one side panel and a command palette: clear a site's cache and data, inspect what a page ships (tech stack, SEO, JSON-LD, headers, Web Vitals, errors, links…), rewrite requests, a dozen utilities, and AI analysis of the page with your own model through llm-connect and llm-models.

Every panel that finds a problem has Copy for agent: a Markdown report a coding agent (Claude Code, Cursor, Copilot…) can act on, with what is wrong, where (selector, JSON path, header, script, cookie name), the exact fix and how to check it. Overview copies one fix list for the whole page. Secrets stay masked and cookie values are left out.

Chrome Web Store · Site: https://maxgfr.github.io/devkit/ (sources in site/, deployed by .github/workflows/pages.yml) · Privacy policy.

Install

From the Chrome Web Store (recommended): click Add to Chrome. Chrome updates devkit on its own.

From a release (no build needed):

  1. Download devkit-<version>-chrome.zip from the latest release and unzip it.
  2. Open chrome://extensions, turn on Developer mode, click Load unpacked and pick the unzipped folder.
  3. To update, unzip the new release over the same folder and click the reload icon on devkit's card.

From source:

pnpm install
pnpm build          # → .output/chrome-mv3, then Load unpacked as above
pnpm dev            # Chrome with the extension and hot reload

Chrome 116 or later.

  • Alt+Shift+D opens the command palette (type json-ld, jwt, fresh…)
  • Alt+Shift+S opens the side panel, Alt+Shift+X clears the site's data and hard reloads
  • Right-click a page or a selection: summarize, explain code, inspect structured data or headers, pick an element, write a bug report
  • Pop out (side panel header) opens devkit in its own window, attached to one tab

Tools

Tab Tools
Cache Storage usage: how much the site keeps on the device (quota, usage by kind, persistent or best-effort), IndexedDB databases with their stores and record counts, Cache Storage caches with entries and sizes, OPFS files, the largest localStorage keys, cookies, service workers, a Markdown report · Hard reload · clear this site's cache · fresh start (all site data + reload) · clear cookies · clear local/session storage · unregister service workers · no-cache mode per tab · site data checklist · cookie editor (edit, import/export JSON, flag audit: Secure, HttpOnly on session cookies, SameSite, expiry, size) · Web Storage editor
Inspect Tech stack (frameworks, CMS, analytics, CDN, server) and DOM stats · SEO & meta (checks, social card preview, headings summary, meta, head links) · Headings outline: every h1–h6 and role="heading" as a tree, with what a screen reader gets (visually hidden, aria-hidden, hidden) and the problems (no h1 or several, headings before the h1, skipped levels shown as gaps, empty or overlong headings, role="heading" without aria-level, duplicates, empty sections), click a heading to find it in the page, level badges drawn on the page, a page map of where the headings fall and how many words each section holds, a table of contents to copy as Markdown or HTML (ids proposed where missing), Copy for agent and AI help (review the outline, rewrite the headings, check them against their sections, find missing sections, or ask about one heading) · Structured data: every JSON-LD block parsed and checked against schema.org and Google rich results (required / recommended properties, ISO dates, absolute URLs, prices, currencies, @id references, breadcrumb positions, JSON syntax errors with line and column), tree and source views, snippet validator, Rich Results Test / Schema validator links · a page report at the top of Overview (security grade, SEO, headings, JSON-LD, errors, accessibility, Web Vitals, each opening its panel) · Performance (Web Vitals on their good / poor scales, navigation timeline, bytes by type, request waterfall, long tasks) · Responsiveness (why it is slow): record while you use the page, then the slowest interactions split into input delay / handlers / paint against INP's thresholds, the scripts that ran in each (Long Animation Frames: function, file, character offset turned into line:column, invoker, forced layout), script time per file, suggestions labeled as heuristics and a copy as Markdown for a coding agent; back/forward cache test with Chrome's not-restored reasons explained; speculation rules (inline and Speculation-Rules header) and whether the page was prerendered or prefetched · Console errors (captured from the first script on every page, count on the toolbar badge) · Link checker · Fonts & colors (eyedropper, palette, contrast) · Accessibility audit with axe-core (WCAG 2.0/2.1/2.2 A and AA, optional best practices and AAA, whole page or one region): violations by impact, needs-review items, each failing element outlined in the page on click, WCAG criteria and Deque's fix guide, Copy for agent / Ask AI · robots.txt tester, sitemap browser, PWA manifest checks · Open in PageSpeed, W3C validator, SSL Labs, securityheaders, Wayback…
Security Headers & security (grade, CSP/HSTS/framing/…, caching, raw headers) · Secrets scanner: API keys, tokens and private keys shipped to the browser (inline scripts and JSON blobs, this origin's scripts, HTML comments, meta tags, config globals, Web Storage, readable cookies), masked until revealed, with where each was found; public-by-design keys (Stripe publishable, Firebase config, Supabase anon) shown as such · Source maps & exposed files: reachable source maps, and an opt-in probe for .git, .env, .DS_Store, phpinfo.php, security.txt… checked by content against a random-path baseline · Third parties & SRI: requests and bytes per site, cross-origin scripts and stylesheets without integrity, crossorigin mismatches, mixed content, iframes without sandbox · Vulnerable libraries: jQuery, jQuery UI, AngularJS, Angular, Bootstrap, Vue, React, Lodash, Moment, Handlebars, DOMPurify… found by their version globals, script URLs and file names (CDN paths included) and the banners of the site's own scripts, matched against a bundled Retire.js database: advisories with severity, CVE / GHSA links and fixed-in version, the first clean release to upgrade to, end-of-life notes (AngularJS 1.x, Vue 2, jQuery 1–2, Bootstrap 3), Copy for agent; detection by version only · CSP analyzer: directive by directive (unsafe-inline, unsafe-eval, wildcards, strict-dynamic, missing object-src/base-uri/frame-ancestors, report-only) and a starter policy from the origins the page used
Page Element picker: click an element, get its React / Vue / Svelte / Angular component stack with file:line (dev builds), props, a unique selector, a Playwright locator, trimmed HTML, the computed styles that matter and a cropped screenshot, then Copy for agent (Markdown for Claude Code, Cursor…) or ask your AI about it · Measure distances: hover for the box model (content, padding, border, margin with px values, size and font), click two elements for the gap between them, Alt for the distance to the parent, drag to measure an area; the panel keeps the last measurement with Copy (CSS values) and Copy for agent (expected value and a note) · Assets: every image, inline SVG, background image, video poster, favicon, manifest icon and social image, with thumbnails, intrinsic vs displayed size ("served 8× larger than displayed"), bytes by type, missing alt text and broken files; open, copy the URL or the SVG markup, download one or all as a zip · design mode · element outlines · disable CSS · block JavaScript per site · emulation (color scheme, reduced motion, contrast, print, CPU and network throttling, vision deficiencies) · responsive sizes · visible / full-page screenshots · form filler
Network Request rules (set/append/remove request or response headers, redirects with regex, block, CORS unblock) with import/export · Mock API responses: fake the body, status, headers or delay of matching fetch/XHR calls, or make them fail like a network error, on the tabs you turn it on for (Chrome debugger, keeps working with the panel closed), a live log of the tab's requests with a count on the panel, a mock from any logged request or from a URL's real response, JSON editor with validation, import/export · environments (local / staging / prod: same path on another env, red ribbon and badge on prod)
AI Bug report for a coding agent (no model needed): your description, an opt-in recorder that turns clicks, typing (values masked), Enter / Escape and navigations into numbered steps and a Playwright test, console errors, failed and slow requests (Resource Timing status), environment (browser, viewport, DPR, color scheme, language), Web Vitals, tech stack, the picked element, a storage summary and optional page text and DOM, as Markdown or JSON with keys and tokens redacted, plus a viewport screenshot to paste; one-click "Copy a bug report" in the palette · Fix plan for the whole page (every finding ranked, each with its fix) · Summarize · explain the stack · SEO audit · heading outline review, heading rewrite, headings vs their sections, missing sections · review/write JSON-LD · security headers review · performance advice · accessibility review · explain console errors · explain selected code · write a Playwright test · extract data as JSON · generate meta tags · UI critique from a screenshot (vision models) · free chat with chosen page data · model browser with prices and context windows
Utils JSON (format, tree, TypeScript types, paths) · JWT · Base64/URL/HTML/hex/unicode · SHA hashes, HMAC, SRI · UUID v4/v7, nanoid · timestamps · regex tester · text diff · colors · cron explainer · URL & query params (strip tracking) · QR code

A DevTools panel ("devkit") lists the page's requests with copy as cURL / fetch / HTTPie / Python, response bodies as JSON trees and HAR export.

AI

Settings (AI tab or the options page) use llm-connect's settings store: any provider of the llm-models catalog (OpenAI, Anthropic, Google, Mistral and every OpenAI-compatible one), or a local server (Ollama, LM Studio, Jan). The key is encrypted at rest with secureStorage (AES-GCM under a non-extractable key) and sent only to the provider. Calls stream from the side panel with callOptions / callProblem / resolveCapabilities (effort levels, web search when the model has it); limits and prices come from llm-models. Page data is trimmed to the model's context window and treated as untrusted in the system prompt.

Ollama refuses unknown origins: start it with OLLAMA_ORIGINS=chrome-extension://*.

public/models.json is the offline fallback catalog; refresh it with pnpm catalog.

Vulnerable libraries database

public/vuln-db.json is a trimmed snapshot of the Retire.js vulnerability repository (repository/jsrepository.json, Copyright 2018 Erlend Oftedal, Apache License 2.0); the commit and date it comes from are in the file and shown in the panel. Refresh it with pnpm vuln-db (scripts/vuln-db.mjs): it drops file hashes, keeps patterns, ranges, severities, CVE / GHSA ids and summaries, and turns Retire's func extractors into plain property paths that the page script reads (no expression from the data is ever evaluated). The Apache-2.0 license, the Retire.js notice and the list of changes ship next to it in public/vuln-db.LICENSE.txt. Matching runs locally; nothing about the page is sent anywhere.

Permissions

Asked at install: activeTab, scripting, storage, sidePanel, contextMenus, and declarativeNetRequest and debugger (Chrome does not allow these two to be optional). browsingData, cookies, contentSettings and host access ("this site" or "all sites") are optional: devkit asks the first time a tool needs them. The options page lists and revokes them.

Development

pnpm check   # biome + tsc + vitest + build
pnpm e2e     # after a build: loads the extension in Chromium and runs every tool against a local test site

src/lib holds framework-free logic (analyzers, rules, utilities, AI context), tested in tests/. src/lib/page holds functions injected into pages: they must not use their enclosing scope. src/ui is the React UI; src/lib/actions.ts and src/lib/panels.ts register every tool for the palette, the side panel and the shortcuts.

Third-party code

devkit bundles axe-core (© Deque Systems, MPL-2.0), unmodified, in axe-audit.js: it is injected only when an accessibility audit runs. Its source is on GitHub and npm. Other dependencies are listed in package.json.

Releases

Releases are automatic. On every push to main, CI runs lint, types, unit tests, the build and the end-to-end test, then scripts/next-version.ts reads the Conventional Commits since the last v* tag:

Commits since the last release Release
fix: or perf: patch (0.1.0 → 0.1.1)
feat: minor (0.1.1 → 0.2.0)
feat!:, fix!: or a BREAKING CHANGE: footer major (minor before 1.0)
only docs:, ci:, chore:, test:, refactor:, style: none

A release builds the zip with that version, publishes it as a GitHub release with generated notes, and, once the Chrome Web Store variables and secrets are set, uploads it to the store and submits it for review. The version lives in the git tags; package.json is not bumped. Store listing texts, images and the API setup are in store/.

About

Chrome extension for web developers: cache & site data, page inspection (JSON-LD, headers, Web Vitals…), request rules, utilities and AI analysis with your own model

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages

⌂ Proxy