SOPs are dead weight.. they live in a Notion folder no one opens. Senior operators document a workflow, junior reps skim it, then everyone goes back to asking the same questions in Slack.

Claude Code skills change that pattern.

A skill is a versioned playbook stored in your GitHub repo. Anyone on the team types one command and the skill runs the same workflow the senior operator built, against the same MCP-connected tools they already use.

That is the shift behind a Claude Code skill library. Your company's playbooks stop living in a doc and start running as agents inside the terminal.

This blog breaks down the five Claude Code skills we use most at workflows.io across ICP modeling, GTM strategy, outbound copy, LinkedIn content, and discovery prep. Every skill below is open-sourced in our Company OS Starter Kit on GitHub so you can clone the file, customize it for your team, and ship.

What a Claude Code skill actually is

A Claude Code skill is a Markdown file with structured instructions that Claude follows every time you call it. The file lives in your GitHub repo. Claude pulls it into the session at start. The instructions specify the workflow Claude should run and the tools it should call along the way.

A skill is different from a prompt. A prompt is one-shot input you type. A skill is a stored, versioned playbook anyone on the team can run. The team improves it through pull requests over time.

For the full architecture of where these skills sit and how the rest of the system wires together, see our breakdown on how to build a Company OS on GitHub.

Why Claude Code skills beat traditional SOPs

A SOP assumes the human reads it and applies judgment correctly. A Claude Code skill assumes Claude runs it and the human reviews the output.

Skills also stay current automatically. Someone updates the skill, every team session pulls the latest version. SOPs in Notion drift the moment you stop maintaining them.

The third difference is execution. SOPs describe work. Skills do the work. A copy SOP tells you how to write a sequence. A copywriter skill writes the sequence, queries past performance data, and hands you the draft ready for review.

A senior operator authors the skill once. Every junior on the team gets the same quality output without needing the senior on every call. That is the compounding effect of a Claude Code skill library.

The 5 Claude Code skills every GTM team needs

These five Claude Code skills cover the GTM workflows we run most often at workflows.io. Each one is open-sourced in the Company OS Starter Kit GitHub repo so you can clone, customize, and run them in your own Claude Code setup.

1. ICP modeller

Turns vague ICP descriptions into a structured tiered model. Pulls firmographic and technographic signals along with account fit data to define Tier 1, 2, and 3 accounts.

The skill encodes the conversations between sales, marketing, and the founder so the ICP definition does not drift each quarter. When a new vertical or persona emerges, you update the skill instead of rewriting three docs.

For a deeper breakdown of how we approach ICP construction, see our ICP modeling guide.

Open the ICP modeller skill on GitHub

Output: a structured ICP scoring rubric every downstream skill reads as input.

2. GTM strategist

Recommends which GTM plays to run for a given segment. Reads the ICP matrix, current pipeline targets, and recent campaign performance.

This is where teams skip the work. They run the same playbook against every segment. The strategist skill forces the question: what play actually fits this tier, and what is the hypothesized lift over what we ran last quarter?

For the full menu of plays the strategist can pull from, see our GTM workflows library and the GTM flywheel playbook.

Open the GTM strategist skill on GitHub

Output: a ranked list of plays with hypothesized lift over baseline.

3. Outbound copywriter

Writes personalized cold email and LinkedIn sequences using your voice guide and the enriched contact data. References your highest-performing past sequences to favor formats that have converted historically.

Generic AI copy reads like generic AI copy. The skill grounds every draft in your team's voice guide and your campaign performance data so the output reads like a human-written first touch instead of getting deleted on sight.

For the underlying messaging framework, see the cold outreach message playbook and our ultimate guide on outbound in 2026.

Open the outbound copywriter skill on GitHub

Output: full sequence drafts ready for human review.

4. LinkedIn post writer

Drafts LinkedIn posts in your voice across hook, body, and CTA. References your team's voice guide and recent post performance to favor formats that have actually worked for your audience.

Most teams either over-edit AI drafts into something stiff or post the raw output and wonder why engagement drops. The skill targets the middle. It gets you 70% of the way there with formats that have performed before, leaving the human to add the specific story or insight that makes the post yours.

For the full content system this skill plugs into, see the 90-day LinkedIn content plan and the complete LinkedIn growth guide.

Open the LinkedIn post writer skill on GitHub

Output: ready-to-edit LinkedIn drafts with hook, body, and CTA.

5. Discovery prep

Researches accounts and contacts before a sales call so the AE walks in with full context. Pulls company news, recent hires, tech stack signals, and the contact's LinkedIn activity into a one-page brief.

The brief lets the AE skip the "tell me about your company" opener and start the call on a sharper question. That changes the energy of the first call without changing the script.

For how this skill plugs into a larger inbound motion across multiple stakeholders, see the multi-thread inbound leads playbook.

Open the discovery prep skill on GitHub

Output: a one-page discovery brief delivered to the AE before every booked call.

Where Claude Code skill libraries break down

A few failure modes show up when teams start building.

1. Over-engineering the first skill. Teams try to write something that handles every edge case. Three pages of clear instructions beats ten pages of conditionals. Ship the simple version and patch what breaks.
2. Shipping skills with no review. If anyone can edit a skill without a pull request, the library drifts and outputs degrade. Treat skills the way engineering treats production code with proper branch protection and PR review.
3. Shipping skills with no MCP layer. A skill that returns a Markdown plan but cannot act in your tools is a smarter SOP, not an agent. The compounding effect shows up when the skill connects to Apollo, Instantly, HubSpot, and the rest of your stack, and ships the work end to end.

Conclusion

The Company OS on GitHub stores your team's knowledge. The Claude Code skill library is what lets that knowledge run. A repo of context files alone is a documentation system with better version control. A repo plus skills is the operating layer of your team.

Every GTM team has a choice in 2026. Document the work in pages no one reads, or ship it as agents the whole team can run from the terminal.

If you want help mapping a tested skill library to your own stack, browse the GTM workflows library or book a strategy call.