MD files in Claude Code: How GTM teams are giving Claude a brain

Use markdown files to build a persistent GTM brain.

Claude Code’s value across teams massively increases with a persistent context layer: a set of markdown files Claude reads before any prompt. ICP, positioning, campaigns, competitive context — all preloaded and shared across the team.

So Claude doesn’t start from scratch. It starts from context for entire team. That shift turns Claude from a tool into a compounding GTM system.

This edition breaks down how to build that system and what to include in your “Claude brain.”

---

Sales today still relies on manual work: stitching together “Frankenstacks” and toggling between a dozen tabs just to log a single call.

Reevo is breaking the mold. As a vertically integrated Revenue Operating System, Reevo replaces the fragmented maze with a single, AI-native workspace. It captures 100% of your first-party data, from emails to meeting recordings, to help reps find high-intent buyers, automate administrative “garbage,” and generate context-rich outreach that actually sounds human.

No more wasting time on data entry or disconnected tools. It’s one unified engine for the entire revenue cycle. Go Stackless. Learn more at reevo.ai.

Influ2 is hosting a webinar on Tuesday, May 6th, 12–1pm ET on a problem most ABM teams struggle with, but few have solved: every buyer takes a different path to a decision, yet most programs still treat everyone at an account the same way.

Join Influ2 and guest speakers Devin Reed (CEO of The Reeder) and Ed Vander Bush (Senior Manager of ABM at LiveRamp), as they break down the failure points that erode buyer momentum and how Influ2’s newest release, built around contact-level orchestration, overcomes them.

Register here for the session - it’s time to orchestrate around how buyers actually buy.

---

The forgetting problem

Claude Code is exceptional at execution. Give it a clear task, point it at the right inputs, and it works through problems with a level of rigor that used to require a strong analyst or engineer.

But it has a structural limitation most teams hit quickly: it doesn’t remember anything once the session ends.

With a single folder, written in plain English, you can transform your Claude usage.

The gap between ad-hoc prompting and a structured context layer

CLAUDE.md

CLAUDE.md is where this all starts.

It’s a simple markdown file that Claude Code automatically reads at the beginning of every session. Not a prompt. Not a system message. It’s a persistent instruction layer that loads your GTM context before you type a single word.

The easiest way to think about it: it’s the difference between onboarding a consultant who already understands your business vs. briefing one from scratch every morning. Same capability, very different output.

This idea caught on quickly. Tools across the ecosystem adopted similar patterns: Google’s Gemini CLI introduced GEMINI.md, and Cursor introduced .cursorrules. Different names, same underlying concept: give the model persistent context upfront.

The key is to unlock CLAUDE.md as the entry point to your GTM operating system - the place where core context lives, gets updated, and compounds over time.

The full GTM repository: CLAUDE.md as index, context files as strategy, skills as repeatable execution

The GTM repository: More than 1 file

CLAUDE.md is the entry point. But the real advantage shows up in what CLAUDE.md points to.

The highest-performing teams aren’t stopping at a single file. They’re building a GTM repository — a structured set of markdown files that encode their go-to-market strategy in a way Claude can read, reference, and act on. Instead of re-explaining context or rewriting processes, they’re systematizing it once and reusing it everywhere.

Most of these repositories break down into three layers:

1) Context: your GTM strategy layer

This is the foundation. It’s the knowledge base Claude uses to understand your business — what you sell, who you sell to, how you win, and what matters right now. It doesn’t change daily, but it evolves over time as your strategy sharpens.

In practice, this includes your company narrative, ICP definitions, messaging and positioning, competitive context, and a running library of high-signal insights. When this layer is strong, every output starts from alignment instead of guesswork.

2) Skills: your execution layer

Skills are where teams start to see step-function gains.

A skill is a reusable markdown file (typically stored in .claude/skills/) that encodes how a task gets done — the process, the structure, and the expected output. Unlike prompts, which disappear at the end of a session, skills persist. Once defined, they can be applied consistently across the entire team.

This is what turns one-off outputs into repeatable systems. Content audits, campaign analysis, meeting prep, post-call summaries — all standardized, all reusable.

The impact is measurable. B2B marketing teams report cutting time spent on content audits and campaign analysis by up to 75% after implementing a proper skill layer. Sales teams see similar gains, with meeting prep and follow-up workflows reducing call-related admin by more than two-thirds.

3) Workflows: your process layer

If skills define how to execute a task, workflows define when and why it happens.

These are your core operating processes translated into something Claude can follow — outbound playbooks, content production systems, deal review frameworks, QBR prep, onboarding checklists. It’s the way your team actually works, made explicit and executable.

Every workflow you encode here is a form of leverage. It’s a process your team doesn’t have to rethink — and a process your competitors likely haven’t systematized.

The six files that capture your GTM strategy in a form Claude can act on

The 6 files that give Claude Code a brain

1. CLAUDE.md: The master index

This file is the first thing Claude reads every session. It should be scannable in under two minutes. Its job is to orient Claude to your current reality: what week it is, which campaigns are live, and what is the most important thing Claude should know before any session this week.

The failure mode most teams hit is making CLAUDE.md too long. It becomes a second context file instead of an index. The rule is simple: if something needs more than three sentences, it belongs in a dedicated context file that CLAUDE.md points to.

Update this file every Monday morning. Five minutes. It is the only file that needs to change every week, and that five-minute investment keeps the entire repository oriented to your current state.

2. profile.md: Your company foundation

Product description, use cases, deal profiles by segment, reference customer names, and team structure. This is the foundation every other file builds on. When Claude generates a research brief, writes a sequence, or drafts a case study, it reads this file first.

3. icp-definition.md: Buyer precision

This is where most teams reveal they do not actually have an ICP. Writing it down forces the precision that verbal definitions allow you to avoid. Not ‘Series B SaaS companies.’ That is a category, not a definition.

A real ICP definition includes the employee range and the reason for it, technographic signals that indicate fit, organizational signals (does this function exist?), explicit anti-ICP exclusions, and a qualification framework with must-haves separated from red flags.

One pattern consistent across every high-performing team: they maintain an ICP evolution log. What changed? When did it change? Why? After a year, that log tells you more about your market than the current definition does.

4. positioning.md: Your messaging clarity

Core value pillars, what not to say, competitive differentiation, and objection frameworks. Every piece of content Claude generates, every sequence it writes, and every battlecard it builds, draws from this file. Without it, Claude defaults to generic category language.

The most valuable line in any positioning.md is the list of things you do not say. Competitors’ terms you avoid. Category positioning you want to escape. Adjectives that have lost their meaning. Claude follows these rules with perfect consistency in a way that no human writer ever manages.

5. competitor-radar.md: Living Intelligence

Win/loss patterns by segment. Common objections they raise. Where you beat them. Where you lose and why. This file is updated after every deal, which means it compounds. The team that has been running this file for six months has a competitive intelligence layer that no static battlecard can match.

Claude Code can automate much of the maintenance. One practical workflow: point Claude at a competitor’s pricing page every Monday and ask it to compare this week’s version against last week’s saved version. If anything changed, write a summary and save it to competitive-intel/pricing-changes.md. Stale context is almost worse than no context. It generates confident but wrong output.

6. signal-library.md: Your Intent Engine

Signals with scoring, detection methods, hook angles that convert, and a performance log per campaign. This is the file that turns Claude Code from a content tool into a revenue tool.

The signal library starts as a hypothesis document. After three or four campaigns, it becomes a learning system. Which signals generated the highest reply rates? Which hook angles drove responses? Which touches actually resulted in meetings? The performance log is the step most teams skip, and it is exactly why their signal library never becomes more than a list of guesses.

How skills compound on context: each layer makes the one above it sharper

Claude Code reached $2.5 billion in annualized revenue by February 2026, with weekly active users doubling in the same period. It overtook GitHub Copilot as the most-used AI coding tool within eight months of launch, and 70% of Fortune 100 companies now use Claude in some form - wild.

How to keep the brain sharp

The most common failure mode teams hit after building a repository is letting it go stale. Context files that do not get updated stop being strategic assets and start being noise sources. Stale context is worse than no context: it produces confident but wrong output.

The teams seeing durable results treat repository maintenance as a recurring operational practice, not a one-time setup task.

Less than 2 hours a week keeps the GTM brain sharp, self-correcting, and compounding

The total ongoing time investment is under two hours a week. The question is whether those two hours get spent or not. Teams that maintain the repository have a self-correcting system. Teams that do not, are running on stale context, which is almost worse than no context at all.

How to build yours this week

The practical starting point is not the most sophisticated architecture. It is the minimum viable repository that actually gets used.

Day 1-2: Write your CLAUDE.md and profile.md

Do not start from a blank page. Ask Claude to interview you about your company, your product, your customers, and your current priorities. Use the output as a draft. Spend 30 minutes refining it. This is the foundation that everything else references.

Day 3-4: Write your icp-definition.md and positioning.md

Force yourself to be specific. Concrete employee ranges. Named technographic signals. An explicit list of anti-ICP exclusions. Things you do not say. If you cannot write these with precision, your GTM motion is running on assumptions that need to be tested anyway. Writing the file forces the reckoning.

Day 5: Build your first skill

Choose the task your team repeats most often. Account research before outreach calls. Post-call summaries. Content briefs. Write a SKILL.md that encodes exactly how you want that task done: the format, the tone, the sources to reference, the output structure. Invoke it with /skill-name. Refine based on output quality. This is the proof of concept that gets the rest of the team bought in.

Week 2: Connect to your live data

MCP servers are now available for most CRMs, analytics platforms, and sending tools. Start with read-only access. Test queries against live data. The first time Claude surfaces a genuinely useful insight from your live CRM data without a manual data export, the value of the full stack becomes clear to everyone watching.

Distribution is still the moat

There is a version of this story that sounds purely technical. Markdown files. Directory structures. Skill invocation syntax. That framing misses the point.

What GTM teams are actually building when they build a repository is a distribution advantage. Every skill file is a process your competitor has not yet made machine-executable. Every signal performance log is a compounding learning loop your competitor is still running manually. Every ICP evolution entry is a market insight that your competitor is either guessing at or not tracking at all.

When Claude Cowork launched in January 2026, the GTMnow community described it as the moment Claude Code became accessible to non-technical GTM operators. That was true. But accessibility was not the insight. The insight was the GTM lesson underneath the Cowork story: when software creation compresses, distribution becomes the constraint.

The same principle applies here. When AI execution compresses, context becomes the constraint. The teams that own the context layer own the output advantage. The teams that do not are prompting from scratch every morning, wondering why their Claude code output does not match their strategic reality.

Your CLAUDE.md is a competitive moat in markdown form.

---

Tag @GTMnow so we can see your takeaways and help amplify them.

---

SpaceX secured an option to acquire Cursor at a $60B valuation — this reframes how the market is pricing AI tools that drive step-function productivity gains. The infrastructure layer of the AI economy is getting consolidated early.

The GTM Repository guide from GTM Strategist goes deep on the ICP evolution log pattern and the signal library maintenance loop. If you are serious about building this over the long term, the section on automating the update loop is worth the read.

GTM 187: How Figma Scaled From $2M to IPO | Kyle Parrish (First Sales Hire)

Listen through the links in the page above or by searching wherever you get your podcasts “The GTMnow Podcast.”

Monk - just raised a $25M Series A (Footwork + Acrew Capital) to fix B2B collections, the $3T sitting in U.S. accounts receivable that takes 45-90 days to actually land. Most AR tools automate the easy stuff, but Monk is built for the edge cases. Customers are seeing 40% lower DSO and 25+ hours saved per month.

SyncGTM - is building the definitive resource for GTM teams running Claude Code in production. Their benchmarks of GTM skill repos (testing ColdIQ, Extruct, and the Claude GTM Plugin against real outbound workflows) are quite rigorous public evaluations of this stack.

1. Enterprise Account Executive at Centari (Hybrid - New York, NY)

2. Enterprise Account Executive, US & Canada at Semrush (Dallas, TX)

3. Senior Manager, Customer Enablement at CaptivateIQ (Remote - US)

4. Enterprise Account Executive at Cube (Hybrid - New York, NY)

5. Account Executive at Obvio (College Park, MD)

6. Founding Account Manager, Accountant at Kick (Toronto, Canada)

See more top GTM jobs on the GTMfund Job Board.

Upcoming events you won’t want to miss:

• Forrester B2B Summit: April 26–29, 2026 (Phoenix, AZ)

• SaaStr Annual: May 12–14, 2026 (San Mateo, CA)

• GTMfund Dinner: May 14, 2026 (San Francisco, CA)

• RevStar Summit: June 3, 2026 (Toronto, Canada)

• GTMfund Dinner: June 9, 2026 (London, UK)

• Dreamforce 2026: September 15–17, 2026 (San Francisco, CA)

• INBOUND: September 16–18, 2026 (Boston, MA)

• Pavilion GTM2026: September 28–October 1, 2026 (NYC, NY)

• CVC Week by Counterpart Ventures: September 29, 2026 (San Francisco, CA)

• Customer Success Week: October 5-9, 2026 (NYC, NY)

• TechCrunch DISRUPT: October 13–15, 2026 (San Francisco, CA)

Some GTMnow Network love to close it out - we appreciate you.

---

---

This newsletter was written and edited by Sophie Buonassisi and the GTMfund team (not AI!).