research-storage_skill

Overview

This skill defines conventions and templates for storing, locating, and retrieving research files produced by dokhak agents. It standardizes directory structure, file formats, and naming to enable reliable caching and multi-tier lookup of research for init and section workflows. Use it to avoid duplicate web searches and ensure consistent file generation and discovery.

How this skill works

The skill provides normalization functions (normalizeChapter, normalizeSection, generateSlug) and a buildCanonicalPath helper to create canonical .research paths. A multi-tier resolution algorithm searches for existing research using progressively broader glob patterns (canonical exact, canonical chapter-section wildcard, non-padded chapter, and flexible keyword match) and returns a resolution object indicating path, existence, and match tier. File templates for summary.md, sources.md, and research.md enforce consistent content structure.

When to use it

• When saving research results from research-collector or researcher agents to ensure canonical paths and templates.
• When checking whether research exists for a chapter/section to avoid duplicating work.
• When reading cached research files for writer or structure-designer agents (read-only).
• When migrating or resolving legacy/inconsistent research directories created with non-canonical names.
• When generating paths programmatically to write or reference research files.

Best practices

• Always run normalizeChapter, normalizeSection, and generateSlug before building paths.
• Use the multi-tier directory resolution algorithm — stop at the first successful match.
• End glob patterns with filenames (e.g., /research.md) because glob does not return directories.
• Write files using the provided templates (summary.md, sources.md, research.md) and include Generated date and sources.
• Consumers should read research files only; do not modify cached research from reader agents.

Example use cases

• research-collector generates .research/init/summary.md and .research/init/sources.md after a domain sweep.
• researcher resolves a section path, appends missing subtopics to research.md, and writes updated sources.md.
• writer reads .research/sections/01-2-core-concepts/research.md to draft page content without re-searching.
• structure-designer checks for existing research using Tier 2/3 patterns to handle legacy directory names.
• Automation script normalizes inputs and creates new canonical directory when resolution returns tier 'new'.