Skilltree

Skilltree

A dead-simple starter kit for non-technical marketers who want AI to work from reusable context instead of a blank chat window.

Skilltree/ architectureRaw work becomes durable context. The system compounds.REPO-NATIVE · HUMAN-IN-THE-LOOPRAWnotes · transcripts · docsMEMORYroots · context storePLAYBOOKStrunk · operationsSKILLSbranches · capabilitiesRUNSfruit · outputsNotesTranscriptsDocsIdeasMeeting takeawaysVoiceStoriesAccountsGlossaryDecisionsGTM launchWeekly newsletterResearch briefCustomer replyDraftSummarizeResearchExtractA rundraft · action · workflow+ MANY MOREJOB 01 · GATHERSReapHarvests raw notesand transcripts. Turnsthem into memory.JOB 02 · CLEANSPruneDedupes memory.Retires stale playbooks.JOB 03 · UPGRADESGraftEvolves memory,playbooks, and skillsas work compounds.HUMAN OPERATORReviews upgrades. Approves skill changes. Sets direction.SKILLTREE · FIGURE 01
Skilltree/ architectureRaw work becomes durable context.The system compounds.RAWnotes · transcripts · docsNotesTranscriptsDocs+2 MOREMEMORYroots · context storeVoiceStoriesAccounts+2 MOREPLAYBOOKStrunk · operationsGTM launchNewsletterResearch+1 MORESKILLSbranches · capabilitiesDraftSummarizeResearch+1 MOREA rundraft · action · workflow+ MANY MOREROUTINESkeep the tree aliveGATHERSReapRaw → memory.CLEANSPruneRetires stale.UPGRADESGraftEvolves the system.HUMAN OPERATORReviews. Approves. Sets direction.SKILLTREE · FIGURE 01
The Skilltree architecture: raw work becomes durable context, which feeds playbooks, which compile into skills, which produce runs.

If you can use a spreadsheet, you can use this.

Why this exists

Most AI workflows fail for a boring reason: the model starts from zero every time.

You open ChatGPT, Claude, or Cursor, paste the same background again, ask for a draft, fix the output, and do it again tomorrow.

The model is usually not the problem. The missing piece is a simple place to keep your context so the tool can reuse it.

That is what Skilltree is for.

It is a starter kit for people who are not especially technical but still want AI to work from real context.

If you can manage a spreadsheet, you can use this.

Why a repo at all?

This is not about becoming a developer.

It is about putting your context in a format AI tools are already good at reading:

  • plain text
  • clear folders
  • reusable files
  • lightweight version history when you want it

GitHub is just the easiest universal home for that right now. It works across tools, keeps the files together, and gives you one place to come back to.

For most marketers, the problem is not "I need better prompts."

The problem is "the useful context lives in five places and the model never sees the same setup twice."

What ships today

Skilltree is intentionally small.

Today it ships with one complete example:

  • a filled example persona
  • a writing-style file
  • a social posting playbook
  • a LinkedIn writing skill
  • one raw-notes file where messy inputs become drafts

The point is not to give you twenty half-finished workflows.

The point is to give you one workflow that makes the idea click.

Copy-paste repo tree

This is the shape:

skilltree/
  AGENTS.md
  context-routing.md
  reference/
    voice/
      writing-style.md
    socials/
      example/
      template/
  playbooks/
    socials/
      writing-a-linkedin-post.md
  skills/
    content/
      linkedin-posts.md
  projects/
    social-content/
      raw-notes.md

That separation matters because the agent should not have to guess what each file is for.

reference/ is durable context.

playbooks/ explain the workflow in plain language.

skills/ tell the AI how to execute that workflow.

projects/ hold live work in progress.

The first workflow: social posting

The first workflow is a LinkedIn draft flow for marketers and operators.

Here is the simple version:

  1. Keep your voice and story context in reference/socials/example/.
  2. Paste rough notes into projects/social-content/raw-notes.md.
  3. Ask the agent to read the playbook, skill, persona files, and notes together.
  4. Get two draft options and a recommendation.

That is enough to feel the difference between:

  • a blank prompt
  • a prompt with reusable context behind it

Five-minute setup

You do not need the terminal for this.

Easiest path

  1. Download the repo as a ZIP.
  2. Unzip it.
  3. Open the folder in Claude Code, Codex, or Cursor.
  4. Open the start-here guide and prompt pack inside the repo.
  5. Paste the setup prompt into your tool.
  6. Let the agent explain the repo and help you replace the example context with your own.
  7. Then paste the draft prompt:
Read AGENTS.md.

Then read:

1. reference/voice/writing-style.md
2. playbooks/socials/writing-a-linkedin-post.md
3. skills/content/linkedin-posts.md
4. reference/socials/example/profile.md
5. reference/socials/example/voice.md
6. reference/socials/example/stories.md
7. projects/social-content/raw-notes.md

Draft 2 LinkedIn post options and recommend one.

If an important detail is missing, ask me 1 or 2 short questions instead of guessing.

The repo now includes a GETTING_STARTED.md guide and a PROMPTS.md file so the user does not have to guess the install path.

What this is not

This is not a polished platform.

It is not trying to automate your whole marketing org on day one.

It is a starter kit.

One working workflow. One clean explanation. One place to start.

Why I think this matters

Most teams already have the raw material.

  • customer notes
  • webinar notes
  • campaign learnings
  • launch retros
  • messages that explain what really landed

What they do not have is one small system that keeps that material reusable.

Skilltree is meant to be that system.

Open the repo

Skilltree on GitHub

Start here inside the repo:

This version is intentionally WIP. The main thing I wanted to get right first was the setup path and the explanation.