YEMPIK / STUDIOSoftware, projects and training.

Yempik Skills: the instructions we use with our AI agents.

A skill is a file of instructions an AI agent reads on its own when the task matches. These are the ones we use in our own work: to verify code, check its security, write down what people know and organise sales work. They are public and you can read every line.

ON THIS PAGEFOR SOFTWARE DEVELOPERSCOMPANY KNOWLEDGE AND AGENTSSTRATEGY AND SALES WORKUTILITIES
The repository on GitHubHow to install them

Written and maintained by Raffaele Zarrelli, Simone Bova and Yempik.

  1. The skillswritten instructions, in 4 groups
  2. Your AI agentreads the one it needs
  3. A personreviews and decides
  1. The skillswritten instructions, in 4 groups
  2. Your AI agentreads the one it needs
  3. A personreviews and decides

WORKS WITH

  • Claude Code
  • Claude Cowork
  • Codex
  • Cursor

And other agents that read skills from a folder. If a skill needs more, its card says so.

FOR SOFTWARE DEVELOPERS

  • verification

    Verification before “done”

    Before saying it has finished, the agent must prove the code works: build, tests, browser, database, logs. And it says what it did not verify.

    We write about it in: The vibe coder’s quote

    Read the code
  • yempik-security-audit

    Code security audit

    A problem is confirmed only when a second agent that did not find it fails to disprove it. It also checks the AI supply chain and ends with one page for management.

    Needs an agent with sub-agents and Node.js. MIT licence, derived from Cloudflare’s security skill.

    We write about it in: The AI supply chain is the new security perimeter · From patch to attack in a few hours

    Read the codeThe audit page
  • screen-recap

    Recorded demo at the end

    Records a test of the result and delivers a short video with a recap, instead of a wall of text.

    Claude Code on macOS only.

    Read the code

COMPANY KNOWLEDGE AND AGENTS

  • knowledge-transfer

    Knowledge transfer

    Interviews a person, one question at a time, and writes down what was never written: exceptions, decision criteria, unwritten rules.

    We write about it in: From tacit knowledge to the agent · How to build a company brain on files

    Read the code
  • agent-reliability-audit

    Reliability of internal agents

    Checks whether an internal AI agent uses current, approved, sourced knowledge: 15 checks, findings ranked by severity, cited evidence. Sources stay read-only.

    We write about it in: How to put an AI agent into production · AI governance for CTOs

    Read the code
  • yempik-company-brain

    Company Brain in the assistant

    The assistant consults the Company Brain before answering from memory and records decisions and rules as proposals to approve, never as truth.

    Needs an MCP connection (the protocol that connects AI agents to other tools and data) to a Yempik Company Brain.

    We write about it in: The Yempik case

    Read the codeThe Company Brain
  • yempik-founder-mode

    Contrarian ideas, verified

    Proposes ideas outside the consensus and checks every claim against the Company Brain’s decisions and facts; whatever is not covered is flagged as such.

    Needs an MCP connection to a Yempik Company Brain.

    Read the code

STRATEGY AND SALES WORK

  • senior-strategy-architect

    Strategy without generic lists

    Turns a vague request (“give me a strategy”) into a plan: the real bottleneck, explicit choices, metrics and risks.

    Read the code
  • gtm-engineering-outbound

    Outbound sales system

    Designs and reviews how you reach new clients: ideal client profile, signals, lists, email and LinkedIn, inbox delivery, reply handling.

    Read the code
  • clay-prospecting

    Prospecting with Clay

    Operating instructions for Clay: targeted search, data enrichment, credit control, excluding contacts already in progress.

    Needs an MCP connection to Clay.

    Read the code
  • yempik-gtm-engineer

    Yempik’s sales method

    Ideal client profile, evidence-based research, drafts grounded in citable facts, weekly review. It only prepares drafts: it never sends anything on its own.

    Needs an MCP connection to a Yempik Company Brain.

    Read the code

UTILITIES

  • audio-video-transcript

    Transcription on your computer

    Transcribes audio and video files on your computer, without sending them to an external service, and produces a recap with a timestamped transcript.

    Needs ffmpeg.

    We write about it in: How to build a company brain on files

    Read the code

How to install them.

  1. With one command

    Adds the skills to your agent. For a single one, add --skill and its name.

    npx skills add yempik-ai/yempik-skills
  2. By hand

    Copy the skill’s folder into your agent’s skills folder; in Claude Code it is .claude/skills/.

  3. The Company Brain skills

    They install as a Claude Code plugin, together with the connection to the Company Brain.

    /plugin marketplace add yempik-ai/yempik-skills

LICENCE

The skills are public: you can read every instruction before your agent follows it.

The repository licence is source-available: you may read, use and modify them for your company, but not redistribute them. The security skill is MIT-licensed.

The repository also contains an SEO skill by another author, under its own licence: it is not listed here because we did not write it.

Want instructions like these for your agents?

Skills are how we write down the way we work. If you want to do the same with your company’s processes, we start from one.

Discuss the project