GitHub PR：web_fetch 增加 `Accept: text/markdown` 以降低抓取 token 成本

通过优先请求 markdown 响应并直返文本，减少 HTML 解析路径的 token 消耗，适合高频网页研究任务。

GITHUBDiscovered 2026-02-13Author echoVic

Prerequisites

OpenClaw runtime supports updated `web_fetch` header behavior (PR merged build).
Target site may provide `Content-Type: text/markdown` (e.g., Cloudflare Markdown for Agents).

Steps

Call `web_fetch` against markdown-capable site and inspect response content-type.
When response is markdown, bypass HTML extraction and return markdown directly.
Benchmark token usage before/after header update on identical URLs.
Keep compatibility fallback: if site ignores Accept preference, continue normal HTML extraction.

Commands

openclaw gateway status

npm run build

openclaw help

Verify

Markdown-capable sites return direct markdown with reduced token footprint; non-markdown sites remain behavior-compatible.

Caveats

Token reduction ratio varies by site template and markdown quality; quoted 80% is scenario-dependent（需验证）.
Some markdown responses may omit embedded data present in raw HTML (scripts/meta blocks).

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.