本文へ移動
cccskills
無料GitHub で公開

nw-jtbd-analysis

JTBD methodology for extracting real jobs behind feature requests — job statements, abstraction layers, first-principles extraction, ODI outcome statements, and opportunity scoring

インストール方法を見る

含まれるファイル(1)

  • SKILL.md5.3 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

JTBD Analysis

Core Principle

A job is the progress a person is trying to make in a particular circumstance. Jobs are stable over time — technology changes, jobs don't. People hire products to make progress; they fire them when they fail.

The trap: Feature requests describe a proposed solution, not the underlying job. Always extract the job first.


Job Statement Format

When [situation/trigger], I want to [motivation/action], so I can [expected outcome]

Examples (good):

  • "When I'm hosting a virtual networking event, I want to facilitate natural conversations between strangers, so I can create valuable connections that wouldn't happen otherwise."
  • "When I join an event as a participant, I want to quickly find relevant people to talk to, so I can maximize the value of my limited time."

Job Types — Extract All Three

TypeQuestionExample
FunctionalWhat task is the user trying to accomplish?Find someone with complementary expertise
EmotionalHow does the user want to feel?Feel confident approaching strangers
SocialHow does the user want to be perceived?Appear professional and well-connected

Abstraction Layers — Navigate to the Real Job

Jobs usually live at strategic or physical level, not tactical.

LayerQuestionExample
TacticalHow do we improve this interaction?Better drag-and-drop for notes
OperationalWhy does this workflow exist?Why do we need a facilitator?
StrategicWhat decision is being pursued?How do we reduce direction uncertainty?
PhysicalWhat's the irreducible function?Input → Synthesis → Convergence

Navigation rules:

  • Use "why?" to move up layers
  • Use "how?" to move down layers
  • Stop when further "why?" produces a life-goal answer

First-Principles Extraction — 3-Step Inversion

When a feature request is presented as a job, apply this:

  1. Identify the Activity — What visible action is the user performing?

    • e.g., "User brainstorms on a whiteboard"
  2. Reject the Activity as the Job — Does anyone wake up wanting to do this activity?

    • "Nobody wakes up wanting to brainstorm" → the job is deeper
  3. Strip to Irreducible Function — What remains if all tools and methods are removed?

    • e.g., "Reduce uncertainty via input-synthesis-convergence"

Disruption check: Is there a higher-level job that would make this entire job unnecessary?


ODI Outcome Statements — Measurable Success Criteria

Format: [Direction] + [Metric] + [Object] + [Context]

Direction: Always "Minimize" (95% of time). "Maximize" only when more is genuinely better.

Metrics priority:

  1. the time it takes to — speed/efficiency (preferred)
  2. the likelihood of — avoiding occurrences
  3. the likelihood that — avoiding results
  4. the number of — quantity reduction
  5. the effort required to — ease

Good vs bad examples:

BadProblemGood
"I want easy video calls"Ambiguous + solution"Minimize the time it takes to start a conversation with a specific person"
"Manage my network effectively"Vague verb + ambiguous"Minimize the time it takes to identify who can help with a specific need"
"Use breakout rooms to talk privately"Solution embedded"Minimize the likelihood of conversations being overheard by unintended parties"
"Don't miss important people"Negative framing"Minimize the likelihood of failing to connect with relevant attendees"

Forbidden words: easy, reliable, good, better, effective, efficient, manage, handle, deal with. Forbidden patterns: solution references ("using AI", "via the app"), compound statements with "and"/"or", demographics.


Opportunity Scoring

Formula: Score = Importance + Max(0, Importance - Satisfaction)

Where Importance and Satisfaction are surveyed 1-10.

ScoreInterpretation
> 12Under-served — high opportunity
10-12Appropriately served — maintain
< 10Over-served — do not invest

Output format per opportunity:

| Outcome | Importance | Satisfaction | Score | Status |
|---------|------------|--------------|-------|--------|
| Minimize time to find relevant attendees | 9.2 | 4.1 | 14.3 | Under-served |

DIVERGE Output for JTBD Phase

Produce docs/feature/{feature-id}/diverge/job-analysis.md with:

  1. Raw request — verbatim feature/problem statement received
  2. Job extraction — 5 Why chain from tactical to physical/strategic level
  3. Job statements — functional (required) + emotional + social (if identifiable)
  4. Disruption check — Is there a higher-level job this entire job is serving?
  5. Outcome statements — 3-6 measurable ODI-format statements
  6. Opportunity candidates — Which outcomes appear most under-served?

Gate: Job must be at strategic or physical abstraction level. Tactical-level jobs are not acceptable input for brainstorming — elevate first.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

Review dimensions for acceptance test quality - happy path bias, GWT compliance, business language purity, coverage completeness, walking skeleton user-centricity, priority validation, observable behavior assertions, traceability coverage, and walking skeleton boundary proof

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

Detailed 5-phase workflow for creating agents - from requirements analysis through validation and iterative refinement

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

5-layer testing approach for agent validation including adversarial testing, security validation, and prompt injection resistance

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

Architectural style selection decision matrices, trade-off analysis, structural enforcement rules, and combination patterns. Load when choosing or evaluating architecture styles.

日本語の概要は準備中です。原文の説明を表示しています。

nWave-ai/nWave6162026年9月16日 更新

nWave-ai のスキルをすべて見る

このスキルの問題を報告する