Token-Efficient Quote Observations

Let find/quote tools match case-insensitively and return abbreviated start–end spans so evidence fits the context budget without full-page dumps.

When a browsing agent must ground answers in page text, expose find and quote actions whose observations are shaped for context engineering: case-insensitive match (quotes also ignore whitespace), and allow abbreviated spans such as <start>━<end> instead of pasting the entire intervening passage. Answering- phase prompts then list numbered quotes with titles, domains, and extracts — compact working memory for citation rather than another full page load.

This instantiates ACI design principles (compact actions, concise feedback) for evidence gathering alongside lm-oriented web observations. Prefer abbreviation over dumping whole pages when the agent already localized the span; keep quote IDs stable so later turns can cite without re-fetching.


Sources

  • WebGPT (Nakano et al.), Appendix A