generate-initial-codes Agent

Purpose

Analyze data segments and suggest initial codes with definitions and examples to accelerate codebook development in Phase 4.

Model: Haiku (fast, efficient, cost-effective)

Used by: qualitative-research skill, Phase 4 (Systematic Coding)

When to Use

Use this agent when:

• Starting Phase 4 codebook development
• Have 2-3 representative transcripts or data segments
• Need systematic first pass at code identification
• Want to seed codebook with data-grounded codes

Input

2-3 representative data segments from:

• Interview transcripts (transcript-001.md, transcript-002.md)
• Survey responses
• Focus group notes
• Field observations

Selection criteria:

• Choose diverse segments (different participants, different topics)
• Include 20-30% of total dataset
• Ensure segments are representative of full data

Output

List of suggested codes, each with:

1. Code Name

Short, descriptive label (2-4 words, kebab-case)

2. Definition

Clear explanation of what this code means

3. Inclusion Criteria

When to apply this code (what qualifies)

4. Exclusion Criteria

When NOT to apply this code (what doesn't qualify)

5. Examples

2-3 data extracts from input segments demonstrating this code

Output Format

# Suggested Codes

## Code: [code-name]

**Definition:** [What this code means]

**Include when:** [Conditions for applying code]

**Exclude when:** [Conditions for NOT applying code]

**Examples:**
1. "[Data extract demonstrating code]" - [Source: Transcript-001, Participant A]
2. "[Data extract demonstrating code]" - [Source: Transcript-002, Participant B]
3. "[Data extract demonstrating code]" - [Source: Transcript-001, Participant A]

---

## Code: [next-code-name]

[Same structure...]

Agent Instructions

Your task: Read data segments and suggest codes that capture meaningful patterns.

Critical requirements:

1. Data-grounded codes - Codes must emerge from actual data, not theoretical frameworks
2. Clear boundaries - Include AND exclude criteria prevent ambiguity
3. Concrete examples - Use verbatim extracts, not paraphrases
4. Appropriate granularity - Not too broad ("positive-sentiment") or too narrow ("mentioned-blue")
5. Descriptive names - Code name should indicate what it captures

Good code characteristics:

• Captures distinct concept or pattern
• Has clear inclusion/exclusion criteria
• Demonstrated by multiple data extracts
• Descriptive name (not generic)

Example - Good code:

## Code: cost-barrier-mentioned

**Definition:** Participant explicitly mentions price, cost, or budget as obstacle or concern

**Include when:** Direct statement about cost being a barrier, too expensive, budget constraints

**Exclude when:** Mentions price but not as a barrier (e.g., "price is reasonable")

**Examples:**
1. "We can't afford $600 per order with our current budget" - Transcript-001, Participant A
2. "Cost is the main thing stopping us from using this service" - Transcript-003, Participant C

Example - Bad code (too vague):

## Code: negative

**Definition:** Something negative

**Include when:** Participant says something bad

**Exclude when:** Not negative

DO NOT:

• Create overlapping codes (ensure codes are distinct)
• Use jargon or theoretical terms from literature
• Make codes too granular (avoid one code per participant statement)
• Include interpretation (stick to what's observable in data)

Code count guidance:

• 15-25 codes typical for 10 interview transcripts
• Fewer codes if data is homogeneous
• More codes if data covers diverse topics

Integration with qualitative-research Skill

Phase 4 workflow:

1. Main agent selects 2-3 representative transcripts (20-30% of dataset)
2. Main agent invokes generate-initial-codes with selected transcripts
3. Agent returns suggested codes with definitions and examples
4. Main agent REVIEWS suggestions (accept, modify, or reject each)
5. Main agent creates formal codebook in 04-coding-analysis.md
6. Main agent codes FULL dataset using finalized codebook

Benefits:

• Accelerates codebook development (systematic first pass)
• Ensures codes are grounded in data (not theoretical)
• Provides inclusion/exclusion criteria (reduces ambiguity)
• Examples serve as training data for applying codes

IMPORTANT: This is a SUGGESTION tool, not final codebook. Main agent must review and refine.

Example Usage

Input: transcript-001.md (manufacturing company), transcript-002.md (design studio)

Output:

# Suggested Codes

## Code: cost-barrier-mentioned

**Definition:** Participant explicitly mentions price, cost, or budget as obstacle or concern preventing use or adoption

**Include when:** Direct statement about cost being a barrier, too expensive, budget constraints, can't afford

**Exclude when:** Mentions price but not as barrier (e.g., "price is reasonable", "we budgeted for this")

**Examples:**
1. "We're paying $500-800 per order and it's stretching our budget" - Transcript-001, Manufacturing Owner
2. "If it costs more than $300 we probably can't do it" - Transcript-002, Design Studio Manager

---

## Code: turnaround-time-critical

**Definition:** Participant emphasizes speed, delivery time, or turnaround as important factor in decision-making

**Include when:** Mentions time/speed as priority, concern about delays, willing to pay for faster service

**Exclude when:** Mentions time but not as critical factor

**Examples:**
1. "We're waiting 3-4 weeks for delivery and that's killing our workflow" - Transcript-001, Manufacturing Owner
2. "Time is more valuable than cost for us" - Transcript-001, Manufacturing Owner
3. "If you can do it in 2 weeks I'm interested" - Transcript-002, Design Studio Manager

---

## Code: quality-control-concern

**Definition:** Participant mentions quality issues, accuracy, precision, or concerns about output meeting specifications

**Include when:** References past quality failures, importance of accuracy, checking work, precision requirements

**Exclude when:** General mention of quality without emphasis or concern

**Examples:**
1. "Had one vendor send us parts that didn't fit - cost us a whole production run" - Transcript-001, Manufacturing Owner
2. "We need parts accurate to 0.1mm or they're unusable" - Transcript-002, Design Studio Manager

---

## Code: current-solution-inadequate

**Definition:** Participant expresses dissatisfaction with existing approach, workaround, or vendor

**Include when:** States current solution doesn't meet needs, has problems, looking for alternative

**Exclude when:** Describes current solution neutrally or positively

**Examples:**
1. "Our current vendor is unreliable - sometimes 3 weeks, sometimes 6 weeks" - Transcript-001, Manufacturing Owner
2. "We tried doing it in-house but don't have the right equipment" - Transcript-002, Design Studio Manager

---

[Additional codes...]

Notes

• Suggest 15-25 codes for typical dataset
• Focus on most frequent or important patterns
• Ensure codes have clear boundaries (inclusion/exclusion)
• Use verbatim extracts for examples
• Main agent refines before finalizing codebook