bootstrap-marketing

bootstrap-marketing

The on-ramp for hypeman. Drops a repo into the hypeman pattern in ~5 minutes. Run this once per repo, then never again (re-running asks before touching anything).

Always do first: locate the plugin

Templates live at <plugin-root>/templates/ — typically .claude/plugins/hypeman/templates/ in the project that's installing. If templates can't be found, surface clearly — the plugin install is incomplete.

Always do next: check for coexistence with cornerman

Read .claude/project.yaml if it exists. Two cases:

Manifest exists (cornerman is installed): good. You're going to add a marketing: block, not overwrite the file. Confirm schema_version: 1 and proceed. If marketing: already exists, ask whether to refresh (default no) or keep + just refresh template files.

Manifest doesn't exist: create a minimal one with just the schema_version and project.name fields, then add marketing:. If cornerman gets installed later, its /bootstrap-project will populate the engineering blocks; the two coexist by design.

Steps

1. Confirm project basics

If the manifest doesn't have project.name, ask. If it doesn't have project.repo, read git remote -v for origin and offer it as the default.

2. Pick a brand profile

Show the profile list from profiles/*.yaml:

• dev-tool — engineering blog primary, HN + X dev community, code-first hooks (shipped in v0.1)

(v0.2 will add early-stage-saas, consumer-app, b2b-enterprise, personal-brand. Until then, anything that isn't dev-tool falls back to custom and the user fills channel defaults manually.)

Print the detection — if the project has a blog directory or content/posts/, lean dev-tool. Ask to confirm or override.

3. Brand voice — three paths (freely mixable)

This is the part hypeman cares most about getting right. Ask which path(s) the user wants — they can pick more than one and the bootstrap will synthesize across them:

Path A — From scratch (no existing content):

For a brand-new project with nothing shipped yet. Walk through 5–8 short questions:

• What's the brand's posture? (helpful / contrarian / dry / playful / earnest / sharp)
• Reading level? (5th grade / general / specialist / technical-deep)
• Formality? (lowercase / sentence case / title case / mixed)
• Sentence rhythm? (short and punchy / long and considered / mixed by section)
• Emoji policy? (none / rare / occasional / liberal)
• Words you love?
• Words you hate? → goes straight into do_not_say
• Brands whose voice you admire? → links into Path C below

Path B — From existing examples:

For projects that already have shipped content. Ask: "Point me at 3–5 pieces of content that exemplify the voice you want — blog posts, X threads, README sections, anything." Paths or URLs both work. Read them, extract patterns:

• Average sentence length, range, variance
• Vocabulary distribution — formal / casual / technical / colloquial
• Recurring stylistic moves (em-dashes, parentheticals, lists, asides)
• Hook patterns — how each piece opens
• CTA patterns — how each piece closes
• Words that recur naturally (candidates for the voice file's "house vocabulary")
• Words conspicuously absent (candidates for do_not_say)

Write the extraction as marketing/VOICE.md with citations back to the source pieces (the lineage matters; future copywriter runs read the exemplars, not just the rules).

Path C — Anchored to reference brands:

For projects that want to sound like a specific other brand (or a blend). Ask: "Name brands whose voice you want to emulate — 2–4 is the sweet spot. URLs to specific pieces of theirs help a lot." Examples Jesse might name: Linear, Vercel, Stripe, 37signals, Pieter Levels, Ben Thompson, Patrick McKenzie, Matt Levine, Anthropic itself.

For each named brand:

• If a URL is provided, fetch and read the piece directly
• Otherwise, summarize what's well-known about that brand's voice from context (and flag this is from memory, not fresh sample — if accuracy matters, ask for URLs)
• Extract the same patterns Path B extracts

Note the blend, not just the union — "Linear's clarity + Stripe's specificity + 37signals' contrarian edge" is more useful than three disconnected style guides. The synthesis goes into VOICE.md's "Aspirational anchors" section, with the original sources cited.

Paths can mix. A common case: Path A for a few baseline questions, Path C for two reference brands, and as the project ships content, Path B gets re-run on the real artifacts. The voice file evolves.

4. Define the primary ICP

Ask for the primary ICP — one persona, the one that matters most. Walk through:

• Title / role
• Where they spend their attention (channels, communities, newsletters)
• What they're tired of hearing
• What signals you're trustworthy to them
• What CTA they'll act on
• Adjacent personas (so you know who else might engage but isn't primary)

Write to marketing/icps/<slug>.md using templates/ICP.md.tmpl. Set marketing.audiences.primary to the new file's path.

If the user wants to skip ICP definition (early prototype, exploring), allow it — write a marketing/icps/TODO.md placeholder and flag in the bootstrap report that an ICP is the first thing to fill before running strategist.

5. Detect + configure channels

Scan the project for channel indicators:

• Blog: content/posts/, content/blog/, pages/blog/, posts/, _posts/ (Jekyll), src/content/blog/ (Astro) → blog.backend candidates
• Existing newsletter integrations: package.json deps on buttondown, resend, mailchimp, @convertkit/* → email.backend candidates
• Existing social config: .well-known/, OG metadata in next.config.* → hints, not decisions

Ask the user to confirm which channels are enabled (not just detected — enabled means we'll actively publish there). For each enabled channel, ask for:

• Account handle / page ID / list ID
• The env var name for the API token (don't ask for the token itself — just the name)
• Whether approval_required: true (default yes for v0.1)

6. Goals

Ask: "What's the one primary metric that defines success for this project's marketing in the next 90 days?" Then up to 3 secondary metrics. Write to marketing.goals.

If the user doesn't have a primary metric yet, the answer is "we don't know what success looks like" — surface that as the gap to close before any campaign should ship. Default primary_metric: "" and flag it in the bootstrap report.

7. Publish policy

Confirm defaults:

• auto_publish: false — strongly recommend keeping this for v0.1
• require_brand_review: true
• require_legal_check_when: [pricing, claims, security] — let user add or remove

Don't make publishing easier than it should be. The cost of an unsanctioned post is much higher than the cost of an extra approval click.

8. Write files

From <plugin-root>/templates/, render each .tmpl against the answers and write into the project:

• .claude/project.yaml — add marketing: block (don't overwrite siblings)
• marketing/VOICE.md ← synthesized from voice paths chosen
• marketing/STYLE.md ← from template, with brand-specific overrides applied
• marketing/PLAYBOOK.md ← empty + header explaining purpose
• marketing/icps/<slug>.md ← from primary ICP answers
• marketing/campaigns/.gitkeep ← so the directory exists
• marketing/examples/.gitkeep ← for house exemplars
• marketing/competitors.md ← empty + header (competitive-watcher writes here in v0.2)
• CLAUDE.md ← inject the <!-- hypeman:* --> managed section
• .gitignore additions ← append marketing/.metrics-cache/ if not present

Each existing file gets a diff preview + confirm-before-overwrite. The manifest marketing: block is the only write that doesn't prompt (re-running bootstrap means refreshing the block).

9. Report

• The manifest path + values populated
• Voice paths used (A / B / C — any combination)
• Which files were created, skipped, or overwritten
• Gaps the user must fill manually (TODO ICP, empty primary_metric, missing env vars for enabled channels)
• Suggested next steps:
  • git add . && git commit -m "chore: bootstrap hypeman"
  • Try /campaign-file against a real signal to test the loop end-to-end
  • Re-run /voice-define once you have 3–5 shipped pieces to upgrade Path A → Path B

Idempotency

• If .claude/project.yaml already has marketing:, ask whether to refresh or keep + only refresh templates
• If managed CLAUDE.md sections exist, refresh in place via markers
• Project-owned content (VOICE.md, ICP files, PLAYBOOK.md, campaigns/) is never touched without explicit confirm

What this skill doesn't do

• Install API tokens (per-machine setup — only ask for env var names)
• File a campaign (use /campaign-file next)
• Validate that the brand voice synthesis is "correct" — voice is subjective, and the iteration loop is the validation
• Migrate from another marketing tooling system