Setup

Step 1 — Understand the business

Work through these questions in stages. Do not ask all at once — ask each block, wait for answers, then move forward.

Block A — Identity:

"What does your company sell, and who do you sell to?"

Wait for answer, then:

Block B — Voice and positioning:

"How would you describe your brand's tone? (e.g., technical/accessible, formal/casual, bold/understated) Any words or phrases that are on-brand? Anything that's off-brand — jargon, hype words, clichés to avoid?"

Wait for answer, then:

Block C — Competitive landscape:

"Who are your top 2-3 competitors? What do they do well and where do they fall short?"

Wait for answer, then:

Block D — Success criteria:

"What does success look like in 90 days? Is there a specific metric that matters most?"

Use all four blocks to inform the config generation. If the user gives a rich first answer, you may skip Block B-D and infer — but present your inferences for review.

Step 2 — Auto-generate config

Use config.example/config.yaml as the schema template. Based on the user's answer, generate:

config/config.yaml

• Keywords tuned to their domain (product names, technology terms, research areas)
• Job titles that map to their buyer personas (economic buyer, technical buyer, champion)
• ICP tiers ranked by fit (Tier 1 = ideal, Tier 2 = good, Tier 3 = marginal)
• Scoring weights that match signal value for their market
• Enable scanners that are likely to surface signal; disable those that won't

config/gtm-context.md

• Product positioning in 2–3 sentences (what it does, for whom, why now)
• Voice and tone rules: describe brand voice in 3-5 adjectives, on-brand language, off-brand language, channel-specific tone adaptations (email vs LinkedIn vs Slack)
• Target personas with pain points and buying triggers per tier
• Competitors section with strengths/weaknesses
• Anti-examples: "content that represents what the brand should never sound like"

config/templates/email-sequences/

• One draft 3-email sequence per enabled scanner signal type
• Follow the 4-sentence rule for initial outreach; low-friction CTA

config/templates/linkedin-sequences/

• One draft connection request + follow-up per enabled signal type
• Keep connection requests under 300 characters

Create config/ if it does not exist. Write all files before presenting for review.

Step 3 — Present inferences for review

Show a concise summary:

ICP Tiers:     [Tier 1 name] / [Tier 2 name] / [Tier 3 name]
Key signals:   [list of enabled scanners]
Disabled:      [list of disabled scanners and why]
Keywords:      [top 10 keywords]
Buyer titles:  [list]

Ask: "What did I get wrong?"

Step 4 — Iterate on corrections

Apply each correction the user identifies and regenerate the affected section. Re-present the summary. Repeat until the user confirms the config looks correct.

Step 5 — Validate

Run /validate on the generated config before confirming setup is complete. Fix any ERRORs before finishing. Surface WARNs to the user with suggested fixes.

Notes

• Reference config.example/config.yaml for field names, types, and defaults.
• Never copy example values verbatim — always infer values from the user's domain.
• If the user's market has no obvious GitHub or ArXiv signal, disable those scanners and explain why.