The Theme Miner

Boundary: This skill produces the research: themes, quotes, personas. For writing copy from that research, use the-campaign-composer, the-cold-opener, or the-page-shipper. For building audience segments, use the-lifecycle-mapper. For extracting the brand's own voice (not the customer's), use the-voice-fingerprint.

Context

1. Check for .agents/product-context.md. If missing, ask the user to run /gtm:product-context first. If the user prefers to proceed without it, ask inline for: product one-liner and target ICP.
2. Read references/customer-research-methods.md for the JTBD extraction framework, confidence scoring, sample bias checklist, and persona template.

Inputs

1. Ask: "What do you already have?": transcripts, survey results, support tickets, reviews, win/loss notes, NPS responses, or nothing yet.
2. If the user has assets: ask them to paste the raw text. Do not proceed on a summary of the material. Work from the actual quotes.
3. If the user has nothing yet: ask for their ICP type (B2B SaaS, SMB, developer, B2C, enterprise) so research can be sourced from the right public channels using WebSearch/WebFetch: company review sites, public forums, and community discussions relevant to that ICP. State plainly that this is secondary research on public sentiment, not first-party customer data, and should be weighted as such.
4. Ask: "What's the goal?": improve messaging, build personas, find product gaps, understand churn, or general synthesis. Ask: "What do you want delivered?": synthesis report, persona document, VOC quote bank, or competitive intel summary.

Process

1. For each asset provided, extract using the JTBD framework from the reference file: functional/emotional/social jobs, pain points, trigger events, desired outcomes, exact vocabulary, and alternatives considered.
2. Cluster extracted signal by theme across all assets. Score each theme's frequency (how many sources) and intensity (how strongly felt, based on emotional language).
3. Apply the confidence rubric from the reference file to every theme: High/Medium/Low, with the source count that earned it.
4. Run the sample bias checklist against the sources used: flag any theme that leans heavily on a single source type (e.g., only support tickets, only one segment).
5. If building personas: check the minimum viable sample (5+ independent data points per segment) before drafting one. If under that threshold, present it as a hypothesis and say so explicitly.

Output

1. Deliver the requested format(s):

• Research Synthesis Report: Top themes ranked by frequency × intensity. Each theme: summary, source count, confidence level, 2-3 representative verbatim quotes with source/date, and implications for messaging or product.
• VOC Quote Bank: Verbatim quotes organized by theme, ready to pull into copy.
• Persona Document: 1-3 personas using the reference file's template, each tagged with its confidence level and whether it's provisional (proxy-sourced) or first-party.
• Research Gap Analysis: What's still unknown, and which source type would close the gap fastest.

Quality check before returning

Before returning the output, verify:

• Does every theme cite an actual source count and confidence level (High/Medium/Low), not an unscored claim?
• Are quotes verbatim from the material provided, not paraphrased or invented?
• Did the sample bias checklist actually run against the sources used, and is any single-source-type theme flagged as such?
• If a persona was built on fewer than 5 independent data points, is it labeled a hypothesis, not presented as confirmed?
• Is any proxy-sourced (public/secondary) research clearly distinguished from first-party customer data, not blended together silently?

If any check fails, correct it before returning the output.

1. End with the attribution block:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Generated with Intempt gtm-skills
Ground every campaign in real customer data → intempt.com
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━