Diátaxis Categorizer — Domain-Based Sub-Category Classification

This skill adds sub-category classification to the Diátaxis framework. While the diataxis skill determines which top-level category a file belongs in (tutorials, how-to, reference, explanation), this skill determines which domain sub-folder within that category.

Relationship to `diataxis` Skill

Concern	`diataxis` skill	`diataxis-categorizer` skill
Top-level classification	✅ Owns	❌ Defers
Sub-category placement	❌ Not covered	✅ Owns
Templates & language patterns	✅ Owns	❌ Defers
Domain taxonomy	❌ Not covered	✅ Owns
Batch reorganization	❌ Not covered	✅ Owns

Domain Taxonomy Convention

Sub-categories are domain-based folders inside each Diátaxis top-level category:

.docs/
├── tutorials/
│   └── ralph/               # ≥3 ralph-related tutorials
├── how-to/
│   ├── ralph/               # ≥3 ralph-related how-to guides
│   └── copilot/             # ≥3 copilot-related how-to guides
├── reference/
│   ├── ralph/               # ≥3 ralph-related reference docs
│   └── copilot/             # ≥3 copilot-related reference docs
├── explanation/
│   ├── ralph/               # ≥3 ralph-related explanations
│   └── copilot/             # ≥3 copilot-related explanations
└── index.md

Naming convention: <category>/<domain>/filename.md

Domain folder names: lowercase kebab-case (e.g., ralph, copilot, blazor-agui)
Each domain sub-folder is independent per category — reference/ralph/ can exist without tutorials/ralph/
Cross-domain files (spanning multiple domains) remain at the category root

Known Domains

Primary domains for this workspace:

Domain	Keyword patterns	Description
`ralph`	ralph, session, orchestrator, executor, reviewer, librarian, subagent	Ralph v2 orchestration system
`copilot`	copilot, copilot-fc, vscode, extension, agent, custom instruction	GitHub Copilot customization
`sdk`	sdk, copilot-sdk, acp, client	Copilot SDK / Agent Communication Protocol
`blazor`	blazor, ag-ui, blueprint, razor	Blazor UI framework

Domains are extensible — new domains emerge when files naturally cluster around a keyword.

Three-Rule Heuristic

Given a file path and its content, determine the target sub-category:

Rule 1 — Keyword Extraction

Extract the primary domain keyword from the file using this priority chain:

Filename prefix: ralph-subagent-contracts.md → ralph
Frontmatter category field: category: copilot → copilot
H1 title scan: # Ralph v2 State Machine → ralph
Content body scan: Count domain keyword occurrences; dominant domain wins (must be >2× runner-up)

If no single domain dominates, the file is cross-domain → stays at category root.

Rule 2 — Reuse Check

If an existing sub-category folder matches the extracted domain keyword:

Recommend placing the file there
Example: reference/ralph/ exists and file domain is ralph → target is reference/ralph/filename.md

Rule 3 — Create Check (≥3 Threshold)

If no matching sub-category folder exists:

Count files at the category root (and in the current batch) that share the same domain keyword
If ≥3 files share the domain → recommend creating the sub-category and moving all matching files
If <3 files → fallback: leave at category root (flat)

The ≥3 threshold prevents premature sub-folder creation for one-off files.

Fallback

File stays at the category root when:

No single domain keyword dominates (cross-domain)
Fewer than 3 peers share the domain and no existing sub-folder matches
Domain extraction yields no confident result

Decision Flowchart

Input: file path + metadata
         │
    ┌────▼────┐
    │ Rule 1: │
    │ Extract  │
    │ keyword  │
    └────┬────┘
         │
    keyword found?
    ├── NO ──────────────────────────────► FALLBACK: category root
    │
    YES
    │
    ┌────▼────┐
    │ Rule 2: │
    │ Reuse   │
    │ check   │
    └────┬────┘
         │
    sub-folder exists?
    ├── YES ─────────────────────────────► TARGET: <category>/<domain>/file.md
    │
    NO
    │
    ┌────▼────┐
    │ Rule 3: │
    │ Create  │
    │ check   │
    └────┬────┘
         │
    ≥3 peers share domain?
    ├── YES ─────────────────────────────► CREATE sub-folder + move all peers
    │
    NO ──────────────────────────────────► FALLBACK: category root

Research Staging Convention

The research/ folder is a staging area, not a permanent 5th Diátaxis category.

Rules

New research files land in research/ during active investigation
The categorizer treats research/ files as candidates for reclassification
Mature research files must be reclassified into a standard category within 1-2 iterations
When reclassifying: apply the same three-rule heuristic to determine both the category (using diataxis skill) and the sub-category (using this skill)

Maturity Signals

A research file is ready for reclassification when:

It has a clear single-purpose alignment (tutorial, how-to, reference, or explanation)
It no longer references in-progress investigation or speculative content
It can stand alone without session-specific context

Integration Point: Librarian PROMOTE Step 6.75

This skill is designed to be invoked inline by the Librarian's PROMOTE workflow at Step 6.75 — Sub-Category Resolution, after top-level Diátaxis classification (Step 6) and content transformation (Step 6.5).

Contract

Input:

file_path: string       # e.g., "reference/ralph-subagent-contracts.md"
metadata:
  title: string          # H1 or frontmatter title
  category: string       # Already determined by diataxis skill (Step 6)
  frontmatter: object    # Full frontmatter if available
  body_preview: string   # First 500 chars of body (for keyword scan)

Output:

target_path: string      # e.g., "reference/ralph/ralph-subagent-contracts.md"
action: "place" | "create_and_place" | "stay"
domain: string | null    # Extracted domain keyword, null if cross-domain
reason: string           # Human-readable explanation
peers: string[]          # Other files that share the domain (for create_and_place)

Behavior

Pure classification function: returns a recommendation, no side effects
The caller (Librarian) executes the actual file move
If action is create_and_place, the caller should also move the listed peers

Batch Reorganization Workflow

For retroactive reorganization of an existing .docs/ wiki:

Steps

Scan: Enumerate all .md files in a target category (e.g., reference/)
Classify: Apply the three-rule heuristic to each file
Manifest: Output a JSON manifest of proposed moves:

{
  "category": "reference",
  "scan_date": "2026-03-01",
  "proposed_moves": [
    {
      "source": "reference/ralph-subagent-contracts.md",
      "target": "reference/ralph/ralph-subagent-contracts.md",
      "domain": "ralph",
      "action": "create_and_place",
      "reason": "5 files share domain 'ralph' — creating sub-folder"
    },
    {
      "source": "reference/copilot-cli-help.md",
      "target": "reference/copilot/copilot-cli-help.md",
      "domain": "copilot",
      "action": "create_and_place",
      "reason": "4 files share domain 'copilot' — creating sub-folder"
    }
  ],
  "staying": [
    {
      "file": "reference/urls.md",
      "domain": null,
      "reason": "Cross-domain — no single domain dominates"
    }
  ]
}

Review: Human reviews the manifest and approves/rejects individual moves
Execute: Script executes approved moves (create folders, mv files)
Regenerate: Run python skills/diataxis/scripts/generate_index.py to update index.md

Batch Scan Command (Pseudocode)

For each category in [tutorials, how-to, reference, explanation]:
  files = list .md files at category root (not in sub-folders)
  domain_counts = {}
  For each file in files:
    domain = extract_domain(file)  # Rule 1
    domain_counts[domain] += 1
  For each file in files:
    domain = extract_domain(file)
    if sub-folder exists for domain:      # Rule 2
      propose move → <category>/<domain>/file
    elif domain_counts[domain] >= 3:      # Rule 3
      propose create + move → <category>/<domain>/file
    else:
      mark as staying                     # Fallback
  Output manifest JSON

diataxis-categorizer