Skip to content

v0.4.3 — Real browser automation (chromedp)

Choose a tag to compare

View all tags
@a-radwan-20 a-radwan-20 released this 15 Jun 16:00
· 34 commits to main since this release
v0.4.3
90c8998

Real browser automation (chromedp)

internal/browser gains a real, JavaScript-capable browser backend. Previously it was HTTP-only with an InteractiveClient that merely described the click/type/submit it would perform; now those actions drive an actual headless Chrome/Chromium via chromedp — pure Go, no CGO.

What's new

  • browser.ChromeClient — manages one persistent headless-Chrome session over the DevTools Protocol: navigate, click, fill, select, submit, read text, screenshot, and state-preserving re-navigation. Multi-step flows (navigate → fill → click → submit) share live page state.
  • The interactive tools now execute for real. browser.click / browser.type / browser.select / browser.submit perform live actions when a browser is installed. All existing security pre-checks still run in both modes: cross-domain blocking, never-type-into-password-fields, and submit confirmation.
  • browser.read_page (new tool) — loads a URL in the live browser (executing JavaScript) and returns the rendered title + visible text, for JS-heavy pages where browser.navigate (static HTTP fetch) returns little. browser.navigate and browser.extract_links stay HTTP-only and fast.
  • browser.screenshot is now real — returns a base64 PNG data URI of the rendered page when Chrome is available.

Graceful degradation

Chrome/Chromium is an external runtime dependency. ChromeClient detects the binary across platform-standard locations; when none is found, Available() is false, the interactive tools fall back to the pre-v0.4.3 validated-description behaviour, and browser.read_page / browser.screenshot return a clear "install Chrome" error rather than crashing. The single static binary is unchanged — Chrome is only needed at runtime for the live features.

Dependencies & build

Adds github.com/chromedp/chromedp (pure Go, no CGO) and its transitive tree (12 direct deps now). The zero-CGO single-binary commitment is preserved; cross-compiles verified for linux/arm64 and windows/amd64; govulncheck reports no vulnerabilities in the new dependency tree.

Tests

6 live integration tests against real headless Chrome (navigate + title/text, JavaScript rendering, fill+click reflecting on the live DOM, PNG screenshot, interactive live-click, and RenderText). They're gated behind AIBUTLER_BROWSER_TESTS=1 so they run locally / on demand but skip on standard CI (GitHub runners ship Chrome, but cold startup there is too slow to gate on). Full repo go test -race -count=1 ./... passes across all 128 packages.

Also de-flaked a pre-existing competing-consumer bus test that surfaced on this release's CI (it asserted a non-guaranteed fairness property; now asserts the deterministic exactly-once invariant + a robust distribution check).

Full changelog: https://github.com/LumabyteCo/aibutler/blob/main/CHANGELOG.md

Assets 2
Loading