Bravenger

Your reps stop losing deals to stale positioning. Your marketing stops contradicting your sales team. Your new hires ramp on this quarter's playbook, not last year's.

Bravenger is a single source of truth that every AI agent reads before it writes. One brain. Six disciplines. Everything that comes out is grounded in how your company actually sells.

View on GitHub

Built by Page Sands · Last updated April 7, 2026

The problem

Your reps are on calls using last quarter's positioning. Marketing is writing blog posts that contradict what sales is saying in outbound. A new hire ramps on a playbook that was last updated six months ago.

None of this shows up as a line item. It shows up as deals that take longer to close, pipeline that doesn't convert, and forecasts that miss.

The root cause: every team, every agent, every piece of content is working from a different version of the truth.

Without a brain
  • Marketing writes copy from a 6-month-old deck
  • Sales rewrites the pitch every outbound cycle
  • New hires ramp on tribal knowledge
  • You update the ICP and nothing downstream changes
  • Every agent invents its own version of the truth
With Bravenger
  • Every agent reads the same ICP, personas, and messaging
  • Agents refuse to produce content outside your ICP
  • One update cascades to every file that references it
  • New hires get this quarter's playbook on day one
  • Drift check catches misalignment before it ships

The proof

It says no when it should.

We asked the marketing agent to write an announcement targeting enterprise buyers. It refused, cited the rule, and offered three alternatives.

Bravenger agent refusing off-ICP content

It catches what you miss.

We changed one field in the ICP — headcount from 100-500 to 100-1,000. Bravenger found 15 stale references across 10 files:

Cascade check: 15 stale references across 10 files

File                        Stale            Should be
brand-vision.md             100-500          100-1,000
core-positioning.md         100-500          100-1,000
messaging-framework.md      100-500          100-1,000
persona-vp-sales.md         100-500          100-1,000
persona-sales-ops.md        100-500          100-1,000
persona-cto.md              100-500          100-1,000
AGENTS.md                   100-500          100-1,000
marketing-usage.md          100-500          100-1,000
uc-001-forecasting.md       100-500          100-1,000
uc-002-coaching.md          100-500          100-1,000

Update all, update selectively, or skip?

One edit. Fifteen things that would have silently gone stale.

How it works

Bravenger is a structured knowledge base that runs inside Claude Code. Three layers:

LayerContainsPurpose
FoundationICP, personas, use cases, goals, brand visionWho you sell to, what they need, and what matters right now
MessagingPositioning, value props, objections, funnel copyHow you talk about it
GovernanceRead order, citations, voice rules, forbidden wordsKeeps every agent aligned

Six agents, each scoped to a discipline

Each discipline has a canonical prompt in prompts/ with thin slash-command wrappers. Every agent loads the full foundation first, then follows discipline-specific guidelines. Same source of truth, different marching orders.

/marketingBlog posts, email campaigns, social, ads, landing pages
/salesOutbound sequences, discovery prep, objection responses, proposals
/pmmPositioning briefs, competitive intel, battle cards, launches
/productFeature briefs, persona reactions, adoption scoring, roadmap input
/csOnboarding guides, migration playbooks, health checks
/drift-checkAudit any content against the brain for violations

Six commands to evolve the brain

Every update triggers a governance check. Change your ICP and Bravenger shows you the downstream impact, then asks before fixing.

/update-icpFirmographics, pains, buying signals, anti-ICP
/update-personaAdd or edit a persona
/update-messagingPositioning, value props, objection handling
/update-brandVoice, values, approved/forbidden language
/update-use-caseAdd or edit a use case
/update-goalsCompany goals and quarterly priorities

Guardrails that hold

  • Mandatory read order. Every agent reads 11 files in sequence before producing anything — voice first, then audience, then content, then rules.
  • Citation requirements. Every claim traces to a [Source: filename#section] reference.
  • Forbidden language. Words banned in your brand voice are blocked across all agents.
  • Anti-ICP enforcement. Agents refuse to produce content for audiences outside your ICP.
  • Cascading updates. Change one field and Bravenger finds every downstream reference, shows you the impact, and asks before fixing.
  • Drift detection. /drift-check audits any content for alignment violations.

Built-in validation

Four lint scripts catch problems before they reach your agents:

npm run lintRun all checks at once
lint:frontmatterYAML frontmatter matches required schemas
lint:citationsEvery [Source:] citation points to a real file and heading
lint:languageNo forbidden words from your brand voice appear in Brain files

Files API upload

Upload your entire Brain to Anthropic's Files API with one command. Every file gets a persistent ID you can reference from any API call.

ANTHROPIC_API_KEY=sk-... node scripts/upload-brain.js
# --dry-run to preview, --force to re-upload

Who it's for

GTM teams at companies up to 50 people, selling 1-2 products in a single market.

Bravenger works when:

  • You have a defined ICP and positioning (even if it's rough)
  • Multiple people or agents are producing GTM content
  • You've caught your sales team saying something marketing didn't approve
  • You've updated your messaging and found stale copy three months later

It ships with a complete example company (NovaCRM — 3 personas, 3 use cases, full messaging framework) so you can see how every file connects before replacing it with your own.

Get started

No coding required. Fork the repo, replace the example company with your own, and run your first agent.

git clone https://github.com/wpsands/bravenger.git
cd bravenger && npm install
  1. Replace the brain. Swap company-brain/ with your ICP, personas, messaging, and brand voice. Keep the file structure and YAML frontmatter schema.
  2. Validate. Run npm run lint to check frontmatter schemas, citations, and forbidden language.
  3. Run an agent. Open Claude Code in the repo and try:
marketing agent — write a blog post targeting [persona] about [use case]
sales agent — prep me for a discovery call with [prospect type]
pmm agent — write a competitive response to [competitor]
Tip Start small. Brand vision, ICP, goals, 2 personas, 2 use cases, and positioning. That's enough to run your first agent.
View on GitHub Fork this repo