tldr

TLDR

Select a brief overview or detailed chat report only from the criteria below.

Route

Route background, delegated, or repository-file research to research; glossary, change-map, or caller-based orientation to repo-map; and review, audit, advice, tutorials, plans, implementation, or remediation to the matching capability.

Use the explicit target, otherwise the current topic segment. Ask one question only when target ambiguity blocks a faithful response.

Select the Mode

Count cues only when they describe the requested response, not quoted text, source content, or a target name.

• Brief: TL;DR, brief, quick, concise, one-screen, high-level; Chinese: 简短, 简要, 快速, 简洁, 一屏, 高层概览.
• Detailed: explore, deep, in-depth, detailed, comprehensive, thorough, full report, analyze; Chinese: 探索, 深入, 详细, 全面, 彻底, 完整报告, 分析.
• Also select detailed when the user explicitly requests at least three of: key facts; mechanism or structure; domain context; limitations or open questions; complete sources.
• If both match, ask only: “Do you want a brief or detailed response?” Do not choose. If neither matches, use brief.

A bare /tldr is not a brief cue. Target type, size, length, complexity, and Agent judgment never affect the mode.

Gather

• Brief: retrieve only enough for identity, purpose, important structure, and material context.
• Detailed: read a URL target first; prefer primary sources for topics and authoritative, directly relevant files for projects.
• Never turn project understanding into a durable report, exhaustive inventory, audit, or implementation plan.
• Separate facts from inference. Disclose conflicts, uncertainty, inaccessible material, and partial coverage.

Output

Brief

• Start with **TL;DR:** and one to three sentences establishing the whole.
• Add the fewest useful bullets, normally no more than five.
• Keep it to roughly one screen; prioritize structure and relationships.