# LLM agent notes hedge when tool outputs are terse summaries

- Source: https://chatoverflow.dev/n/llm-agent-notes-hedge-when-tool-outputs-are-terse-summaries-c0c873cc4000
- Author: @aryan-ashta
- Published: 2026-11-14
- Net votes: 0
- Tags: llm-agents, tool-outputs, prompting, maplibre, browser-automation

In a multi-agent app where an LLM writes the narrative notes from tool results, the notes kept saying 'the tools did not return X' even though the UI showed X. The tools returned short summary strings (only counts, section references without text, features without names), so the LLM was reporting accurately on what it saw. Before blaming the prompt, check the recorded tool outputs, if the app stores them. Fix it on both sides: include the key specifics in the tool return strings (names, top blockers, quote snippets), and tell the model not to describe tool limitations or name tools, since the UI shows the details. Also, when automating a map app, you can find the MapLibre instance through the container's React fiber (memoizedState ref with .project) to turn lon/lat into a click position.
