mcpField
T
theory-analysis-wom-motivation

theory-analysis-wom-motivation

github.com/timlai666Marketing & growth044 installsUpdated 1d ago
Repository

About theory-analysis-wom-motivation

This skill analyzes why people share experiences or opinions using Word-of-Mouth Motivation Theory. It accepts evidence from interviews, support conversations, social posts, research notes, and observations, and returns traceable annotations in JSON and Markdown.

Key features

  1. 01Checks required input before analysis
  2. 02Codes four word-of-mouth motives
  3. 03Links conclusions to quoted evidence
  4. 04Outputs JSON and Markdown
  5. 05Supports optional review-mining-stp handoff

Use cases

  1. 01Analyzing motives in interview summaries
  2. 02Coding evidence from support conversations
  3. 03Comparing motives across social posts
  4. 04Preparing annotations for review-mining-stp

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/timlai666/skills --skill theory-analysis-wom-motivation

After installing, type/theory-analysis-wom-motivationin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents/
openai.yaml
references/
00-theory-foundation.md
01-input-and-gate.md
02-dimension-coding-rules.md
03-output-contract-json-markdown.md
04-worked-example-turbocharger.md

Details

From the project README

Theory Analysis WOM Motivation

Overview

用口碑動機理論(Word-of-Mouth Motivation Theory)解析證據提供者為何願意分享經驗或態度。 來源可為訪談摘要、客服對話、社群貼文、研究筆記、觀察紀錄等,不限於評論。 本 skill 含完整理論章節,可在分析前先校準口碑動機的判讀邏輯。 此 skill 可單獨使用,也可選擇把結構化結果交給 review-mining-stp。

Input Contract

至少提供:

  • analysis_goal: string
  • evidence_items: array
    • item_id: string
    • content: string
    • content_type: string

建議可選:

  • source_type: string
  • source_ref: string
  • context_tags: string[]

Data Sufficiency Gate

缺必要資訊時,輸出 MissingDataOutput,不直接推論口碑動機。

{
  "missing_fields": [],
  "why_needed": {},
  "questions_to_user": [],
  "next_step_rule": "Provide analysis_goal and at least one evidence item with item_id, content, and content_type, then rerun coding."
}

Workflow

  1. 先讀 references/00-theory-foundation.md,校準理論概念、動機類型與研究應用語境。
  2. 讀 references/01-input-and-gate.md,檢核資料完整度。
  3. 讀 references/02-dimension-coding-rules.md,依四種口碑動機做證據標註:
    • altruistic
    • social_identity
    • self_enhancement
    • emotional_expression
  4. 讀 references/03-output-contract-json-markdown.md,輸出 JSON + Markdown。
  5. 用 references/04-worked-example-turbocharger.md 對齊格式與深度。

STP Integration

  • theory_annotations.family 固定為 wom_motivation
  • theory_annotations.subtheory 只可使用:
    • altruistic
    • social_identity
    • self_enhancement
    • emotional_expression

結果可直接對接 review-mining-stp 的理論標註。 review-mining-stp 屬於 optional downstream,不是本 skill 的必要依賴。

Quality Rules

  • 口碑動機需要有可辨識意圖,若無意圖訊號需標記 insufficient。
  • self_enhancement 與 social_identity 必須區分:
    • 前者偏「展現專業或能力」
    • 後者偏「取得社群認同或歸屬」
  • 所有結論都要可被 evidence_quotes 追溯。

Suggested Prompts

  • Use $theory-analysis-wom-motivation to code sharing motives from mixed evidence items and return JSON + Markdown in Traditional Chinese.
  • 請用 $theory-analysis-wom-motivation 分析多來源證據中的口碑動機,並輸出可選串接 review-mining-stp 的 theory_annotations。

Comments

Markdown supported