#15414: feat(web-fetch): add Accept: text/markdown header for Cloudflare Markdown for Agents

## Summary Adds `Accept: text/markdown, text/html;q=0.9, */*;q=0.8` to `web_fetch` HTTP requests, enabling [Cloudflare Markdown for Agents](https://blog.cloudflare.com/markdown-for-agents/) support. When a server returns `Content-Type: text/markdown`, the body is used directly (extractor: `markdown-native`) — skipping HTML extraction entirely. This can reduce token consumption by ~80% on supporting sites. **Fully backward-compatible:** non-supporting sites return `text/html` as usual and hit the existing Readability/Firecrawl pipeline. ## Changes - `src/agents/tools/web-fetch.ts`: Updated `Accept` header + added `text/markdown` content-type handling ## Testing - TypeScript compiles clean (`tsc --noEmit`) - All 11 existing web-fetch tests pass - Linter (oxlint) + formatter (oxfmt) clean Closes #15393 🤖 AI-assisted (Claude Opus 4) <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This change updates `web_fetch` to send an `Accept: text/markdown, text/html;q=0.9, */*;q=0.8` header and adds a fast-path for `Content-Type: text/markdown` responses (marking them as `extractor="markdown-native"` and skipping the HTML Readability/Firecrawl extraction pipeline). Overall this fits cleanly into the existing fetch/extract flow, but there are a couple of behavioral inconsistencies to fix so `extractMode` remains the primary contract and content-type detection is robust. <h3>Confidence Score: 3/5</h3> - Mostly safe to merge after fixing output-format consistency for markdown responses. - The PR is small and well-scoped, but currently changes `web_fetch` semantics in a way that can surprise callers: `extractMode: "text"` can return markdown when a server replies `text/markdown`, and Content-Type detection is case-sensitive. Both are straightforward to address. - src/agents/tools/web-fetch.ts _{Last reviewed commit: 8de4bb6} <!-- greptile_other_comments_section --> <!-- /greptile_comment -->