mcpbeat Sign in

Nw Divio Framework Agent Skill

DIVIO/Diataxis four-quadrant documentation framework - type definitions, classification decision tree, and signal catalog

2k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
588
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/nWave-ai/nWave --skill nw-divio-framework

The instruction itself

17 sections, as written by the author

DIVIO Documentation Framework

The Four Quadrants

Exactly four documentation types. Each serves one purpose. Never mix.

Tutorial

Orientation: Learning | Need: "Teach me" | Key Q: Can newcomer follow without external context?

Purpose: enable first success | Assumption: user knows nothing | Format: step-by-step guided experience

Success: gains competence + confidence | Include: safe repeatable steps, immediate feedback, building blocks

Exclude: problem-solving, assumed knowledge, comprehensive coverage

How-to Guide

Orientation: Task | Need: "Help me do X" | Key Q: Achieves specific, measurable outcome?

Purpose: accomplish specific objective | Assumption: baseline knowledge, needs goal completion

Format: focused steps to outcome | Success: task completed

Include: clear goal, actionable steps, completion indicator | Exclude: teaching, background, all scenarios

Reference

Orientation: Information | Need: "What is X?" | Key Q: Factually complete and lookup-ready?

Purpose: accurate lookup | Assumption: user knows what to look for | Format: structured, concise, factual

Success: finds correct info quickly | Include: complete API/function details, parameters, returns, errors

Exclude: narrative, tutorials, opinions

Explanation

Orientation: Understanding | Need: "Why is X?" | Key Q: Explains reasoning and context?

Purpose: conceptual understanding | Assumption: user wants "why" | Format: discursive, reasoning-focused

Success: understands design rationale | Include: context, reasoning, alternatives, architectural decisions

Exclude: step-by-step, API details, task completion

Classification Matrix

                  PRACTICAL           THEORETICAL
STUDYING:         Tutorial            Explanation
WORKING:          How-to Guide        Reference

Adjacent: Tutorial/How-to (both have steps, differ in assumed knowledge) | How-to/Reference (both "at work") | Reference/Explanation (both knowledge depth) | Explanation/Tutorial (both "studying")

Classification Decision Tree

START: What is the user's primary need?

1. Is user learning for the first time?
   YES -> TUTORIAL
   NO  -> Continue

2. Is user trying to accomplish a specific task?
   YES -> Does it assume baseline knowledge?
         YES -> HOW-TO GUIDE
         NO  -> TUTORIAL (reclassify)
   NO  -> Continue

3. Is user looking up specific information?
   YES -> Is it factual/lookup content?
         YES -> REFERENCE
         NO  -> Likely EXPLANATION
   NO  -> Continue

4. Is user trying to understand "why"?
   YES -> EXPLANATION
   NO  -> Re-evaluate (content may need restructuring)

Classification Signals

Tutorial Signals

Positive: "Getting started", "Your first...", "Prerequisites: None", "What you'll learn", "Step 1, Step 2...", "You should see..."

Red flags: "Assumes prior knowledge", "If you need to...", "For advanced users..."

How-to Signals

Positive: "How to [verb]", "Before you start" (with prerequisites), "Steps", "Done:" or "Result:"

Red flags: "Let's understand what X is...", "First, let's learn about..."

Reference Signals

Positive: "API", "Parameters", "Returns", "Throws", "Type:", Tables of functions/methods

Red flags: "This is probably...", "You might want to...", Conversational tone

Explanation Signals

Positive: "Why", "Background", "Architecture", "Design decision", "Trade-offs", "Consider", "Because"

Red flags: "1. Create...", "2. Run...", "Step-by-step", "Do this:"

JTBD-First Resolution (When in doubt, ask the JTBD)

Provenance: Ale 2026-05-02. "La JTBD ci dice quale need stiamo soddisfando."

When the four-types decision tree is ambiguous, OR when a stylistic decision inside a chosen type has plausible alternatives (ordering, fold-vs-keep, scratch-vs-real-data, prose-vs-table, etc.), DO NOT pick by aesthetic preference. Return to the document's JTBD and let the answer fall out.

Each Diataxis type maps to a canonical JTBD:

| Type | Canonical JTBD | Implied constraints |

|---|---|---|

| Tutorial | "Build my mental model of X by doing it once, end-to-end, without risking my real project state." | Reproducible, safe, self-contained, cause-effect visible |

| How-to | "Fix this specific situation right now; don't teach me fundamentals." | Problem-statement first, decision tree, no preamble, link forward to Tutorial for newcomers |

| Reference | "Look up a specific answer (flag, exit code, schema field, error message)." | ctrl-F friendly, dense, alphabetical/systematic, no narrative, separate H2 per lookup category |

| Explanation | "Understand the design rationale so I can extend or critique it." | Discursive, links to evidence (probes, RCAs, ADRs), no how-to instructions, no command examples beyond the minimum needed for context |

Application: resolving a stylistic dispute

Worked example, Tutorial Q: scratch directory vs use the project's seeded data?

  • JTBD: "Build mental model end-to-end without risking my real project state."
  • Constraint surfaced: "without risking" → seeded data DOES risk contamination → scratch wins.
  • Constraint surfaced: "end-to-end" → cause-effect visible → seeded data hides cause-effect under pre-existing entries → scratch wins.

Worked example, Reference Q: fold three short sections into one "Notes"?

  • JTBD: "Look up a specific answer."
  • Constraint surfaced: "specific answer" → ctrl-F friendly → 3 distinct H2 entries are 3 distinct landing points → folding hurts the JTBD.
  • Verdict: keep separate.

Worked example, README "Learn More" Q: alphabetical or semantic grouping?

  • JTBD on a README is "I'm new; show me docs in the order I should read them" — pedagogical browsing, NOT lookup.
  • Constraint surfaced: "in the order I should read them" → semantic grouping (install adjacent, authoring adjacent, troubleshooting last) → wins.
  • Alphabetical serves a DIFFERENT JTBD ("I know the doc name") which has another surface (file system / search).

Decision protocol

When facing a stylistic ambiguity inside a chosen type:

  • Restate the document's JTBD in one sentence (use the canonical JTBD as the starting point, refine if the doc is more specific).
  • List the constraints implied by that JTBD (be concrete: "reproducible", "ctrl-F friendly", "links to evidence").
  • Compare each candidate option against the constraint list.
  • Pick the option satisfying the most constraints. If tied, pick the option that fails the FEWEST constraints (loss-aversion).

This procedure replaces aesthetic debate with a falsifiable check.

Anti-pattern: collapse via JTBD-blending

Some authors blend two JTBDs in one document ("a Tutorial that's also a Reference"). This is the same collapse pattern the four-types-only rule rejects, surfaced via the JTBD lens. Detection: if the document has TWO canonical JTBDs in tension (e.g. "build mental model" AND "look up specific answer"), flag for splitting.

Other skills for the same job

different authors, same section of the catalogue
GitHub Project Management
by ComeOnOliver
×3

Comprehensive GitHub project management with swarm-coordinated issue tracking, project board automation, and sprint planning

14k tokens
Folder Structure Blueprint Generator
by github
vendor ×1

Comprehensive technology-agnostic prompt for analyzing and documenting project folder structures. Auto-detects project types (.NET, Java, React, Angular, Python, Node.js, Flutter), generates detailed blueprints with visualization options, naming conventions, file placement patterns, and extension templates for maintaining consistent code organization across diverse technology stacks.

3k tokens
Sequential Thinking
by mrgoonie
×1

Use when complex problems require systematic step-by-step reasoning with ability to revise thoughts, branch into alternative approaches, or dynamically adjust scope. Ideal for multi-stage analysis, design planning, problem decomposition, or tasks with initially unclear scope.

4k tokens
Openserv Multi Agent Workflows
by internet-court
×1

Multi-agent workflow examples to work together on the OpenServ Platform. Covers agent discovery, multi-agent workspaces, task dependencies, and workflow orchestration using the Platform Client. Read reference.md for the full API reference. Read openserv-agent-sdk and openserv-client for building and running agents.

24k tokens
Caveman Compress
by HoangNguyen0403
×1

> Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md.

7k tokens scripts
API Patterns
by lingxling
×1

API design principles and decision-making. REST vs GraphQL vs tRPC selection, response formats, versioning, pagination.

5k tokens scripts
Github Workflow Automation
by lingxling
×1

Patterns for automating GitHub workflows with AI assistance, inspired by [Gemini CLI](https://github.com/google-gemini/gemini-cli) and modern DevOps practices.

5k tokens
Domain Identification Grouping
by christophacham
×1

Groups existing components into logical business domains to plan service-based architecture. Use when asking "which components belong together?", "group these into services", "organize by domain", "component-to-domain mapping", or planning service extraction from an existing codebase. Do NOT use for identifying new domains from scratch (use domain-analysis) or analyzing coupling (use coupling-analysis).

10k tokens

How to use it

Copy the folder

Take nwave-ai/nw-divio-framework from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.