# Analyze First > Enforces a strict analysis-before-synthesis discipline rooted in human-centered design (HCD) methodology. Trigger this skill whenever the user shares research data of any kind — interview notes, transcripts, survey responses, observation notes, JTBD artifacts, usability findings, discovery outputs, or any qualitative or quantitative research material. Also trigger when the user says phrases like "analyze this," "slow down," "let's be methodical," or "don't jump ahead." If the user shares research data WITHOUT specifying a goal or workflow, always pause and ask whether they want analysis or synthesis mode before proceeding. Never skip to synthesis, conclusions, recommendations, ideas, concepts, action items, or next steps without explicit user permission. This skill has two mandatory gates: (1) confirm all data is collected before beginning analysis, and (2) ask permission before moving into synthesis. - Skill: `callahkate/analyze-first` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add callahkate/analyze-first` - Raw SKILL.md: https://api.skillmd.com/api/skills/callahkate/analyze-first/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: callahkate (https://skillmd.com/u/callahkate) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/callahkate/analyze-first --- # Analyze First A skill for enforcing rigorous analysis discipline before any synthesis, interpretation, or ideation. Rooted in human-centered design methodology. --- ## The Core Distinction | | Analysis | Synthesis | |---|---|---| | **Question answered** | "What did we see, hear, or learn?" | "What does it mean when findings are considered together?" | | **Movement** | Breaking DOWN data into observations, facts, details | Building UP pieces into patterns, themes, insights | | **Stays close to** | Evidence | Interpretation and implication | | **Outputs** | Observations, inventories, timelines, frequencies, behaviors, quotes, context | Themes, insights, mental models, design implications, recommendations | **Analysis must always precede synthesis. Synthesis requires explicit user permission.** --- ## Two Mandatory Gates ### Gate 1: Confirm Complete Data Collection Before beginning any analysis, confirm you have everything: - "Before I start analyzing — do you have all your data together, or is there more coming?" - If more data is expected: wait. Do not begin analysis on a partial dataset. - If data is complete: proceed to Gate 2. ### Gate 2: Ask Mode Before Proceeding If the user shares research data without specifying a goal: > "I can see you've shared [type of data]. Before I dig in — are we in **analysis mode** (breaking down what we saw and learned) or **synthesis mode** (building toward meaning, themes, and implications)?" If they say analysis: follow the Analysis Protocol below. If they say synthesis: confirm they're ready to move past analysis, then proceed. If they say they're unsure: recommend analysis first and explain why briefly. --- ## Analysis Protocol ### What Analysis Includes Describe only what is observable in the data. Stay descriptive and factual. **Depending on study type, capture any relevant combination of:** - **Who**: participant roles, customer types, demographics, screening criteria - **What**: specific behaviors, actions, stated needs, tools used, objects involved - **When**: timing, sequence of events, moments in a journey or process - **How much / How often**: frequency, volume, duration, repetition - **Steps involved**: process flows, task sequences, decision points - **Context**: environment, social or business situation, surrounding conditions - **JTBD artifacts** (if applicable): first thought, consideration set, moment of first use, moment of satisfaction, switching triggers, anxieties - **Meta-observations**: interviewer notes on participant energy, affect, body language — when they seemed excited, bored, confused, hesitant, or flustered; pacing; who dominated; what felt rehearsed vs. spontaneous ### What Analysis Excludes Do NOT include any of the following during analysis phase: - ❌ Themes or patterns across participants ("users tend to...") - ❌ Interpretations ("this suggests...") - ❌ Implications ("this means for design...") - ❌ Recommendations, action items, next steps - ❌ Reframing observations as "pain points," "opportunities," or "needs" — these are synthesis moves - ❌ Grouping findings by meaning (grouping by type or participant is fine) ### Output Format for Analysis Structure analysis output to match the study type. Default structure: ``` ## Participant / Data Source: [ID or label] **Role / Context:** [who they are, their situation] **Key Events / Actions:** [what happened, in sequence if relevant] **Stated Details:** [what they said, close to verbatim where useful] **Frequencies / Quantities:** [how often, how much] **Meta-observations:** [affect, energy, notable interviewer notes] ``` Repeat per participant or data source. Do not collapse across participants during analysis. --- ## The Synthesis Gate When analysis is complete, stop fully. Then ask: > "Analysis is complete. I've broken down [X participants / data sources] into observations and details. > > Ready to move into synthesis — where we start building up toward patterns, themes, and what it means? > Or do you want to sit with the analysis first?" Do not proceed until the user explicitly says yes. --- ## Trigger Phrases (always activate this skill) - "analyze this" / "let's analyze" / "analysis first" - "slow down" / "don't jump ahead" / "be methodical" - "what did we learn" / "what did we see" - "let's go through the data" - User shares interview notes, transcripts, survey data, observation notes, or any research artifact WITHOUT specifying a goal ## Anti-patterns to Avoid - Jumping to "this suggests..." before synthesis is unlocked - Grouping findings into themes during analysis phase - Offering recommendations embedded in analysis - Treating "interesting" or "notable" as synthesis-safe words when they imply meaning - Rushing through analysis because data seems "obvious"