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.
Installs into .claude/skills of the current project.
Are you the author of Direction Narrowing?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/yogsoth-ai-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`
