fix(browser): make get_text extraction method and get_html truncation configurable

[danielferreira-dias]

Summary

• get_text: Default text extraction changed from text_content() to inner_text(),
  which excludes script/style/hidden content. The old behavior is available via
  method: "text_content" on GetTextAction.
• get_html: Removed the hard-coded 1,000-character truncation. Full HTML is now returned
  by default. An optional max_length field on GetHtmlAction lets callers opt into truncation.

Motivation

• text_content() includes <script>, <style>, and hidden element text, polluting LLM
  context when agents read pages. inner_text() is style-aware and produces much cleaner output.
• The 1,000-char HTML truncation made get_html unusable for reading full page content,
  and the LLM was never informed about it. Meanwhile get_text had no truncation at all,
  making the limit inconsistent.
• get_html returns proper HTML with tags (<script>, <style>, <nav>, etc.), which
  downstream processing (like markdown conversion) can strip effectively. Without full HTML,
  the only option is get_text where JavaScript/CSS appears as untagged noise that cannot
  be reliably cleaned.

References

• Playwright: innerText vs textContent — text_content() includes scripts/styles, inner_text() excludes them
• Playwright Issue #18894 — edge case where inner_text can still include script content; Playwright team recommends stripping <script> elements as workaround
• Playwright itself imposes no character limit on content(), inner_html(), or text_content() — the 1,000-char truncation was an arbitrary application-level constraint

Test plan

• Verify get_text with default (no method field) uses inner_text and excludes scripts/styles
• Verify get_text with method: "text_content" returns raw text including scripts
• Verify get_html with no max_length returns full HTML without truncation
• Verify get_html with max_length: 500 truncates and appends "..."
• Existing browser tests still pass

[danielferreira-dias]

We have a browser automation agent that navigates pages and processes their content. We built an AfterToolCallEvent hook that intercepts get_html output, strips noise (<script>, <style>, hidden elements, etc.) and converts the HTML to clean markdown using markdownify.

The problem is the 1,000-char hard truncation on get_html — by the time our hook receives the output, it's already an incomplete HTML fragment. This makes the entire get_html → markdown conversion pipeline useless.

Removing the truncation would let downstream hooks and tools properly process full page HTML, which is a cleaner approach than relying on get_text/text_content() that includes unrendered scripts and CSS noise.

[danielferreira-dias]

This would be extremely valuable to fix, at least removing the get_html truncation limitation, which currently makes downstream HTML processing (e.g., HTML-to-markdown conversion via hooks) non-functional.

Feature Issue
Bug Issue