direction-narrowing

---
name: direction-narrowing
description: Focus within the user's chosen field(s). Identify specific sub-directions through deep paper and web research, then present ranked candidates. Use after landscape-reconnaissance has identified fields of interest.
---

# Direction Narrowing

Focus within chosen field(s). Identify specific sub-directions and present ranked candidates.

## Available SOPs

| SOP | Purpose | Execution |
|-----|---------|-----------|
| broad-paper-search | Scan papers in the chosen field(s) | import: literature-overview |
| deep-web-search | Deep reading of web resources in the field | subagent |
| present-candidates | Present ranked sub-directions to user | dialogue |

## Methodology Guidance

- hot-start may only need partial SOP execution (a few searches for context)
- You decide search depth based on information sufficiency
- `present-candidates` depth scales by start mode:
  - cold-start: broad sub-directions available to pursue
  - warm-start: specific sub-problems and research tracks
  - hot-start: granular knowledge points, technical details

## Hard Constraints

- `broad-paper-search`: at least 80 papers scanned
- `deep-web-search`: at least 30 web pages read in full

## Output (Tactic-Level Aggregation)

`RankedCandidates[] + user's selection`