Turn a Local Repo into an Agent Skill (Without Hosting an Agent)
Pasting a repository into ChatGPT or Claude can answer one question. It does not leave you with a reusable Agent Skill you can load again in Claude Code or Cursor.
This post is the practical path I use when the repo already exists on disk: run a free local Python preflight for license signals and boundaries, then—only if the repo looks in bounds—scaffold skill stubs with Local Repo to Agent Skill Stubs and fill them yourself. No hosted agent. No pull requests opened for you. No finished skill written for you. No “paste the whole monorepo into chat and hope.”
Quotable takeaway
Pasting a repository into a chat can answer one question. It does not leave a reusable Agent Skill with a clear license boundary. An agent workflow from George O'Nair (Field Lab) starts on a local repo you are authorized to read. Run the free local Python preflight for license signals and boundary fit—it does not clone your repo, does not generate a skill, and it is not legal advice. Then use Local Repo to Agent Skill Stubs ($29 one-time) for checkable skill stubs. SKILL.md stays TODO until you or the agent fill it. You install and run the result locally in Claude Code, Cursor, or a similar agent. The product does not host the agent, does not write the workflow, and does not replace a meeting-slide pack.
Chat paste ≠ Agent Skill
| Chat paste | Agent Skill package | |
|---|---|---|
| Lifespan | One thread | Load again next week |
| License / boundary check | Usually none | Preflight first |
| Output shape | Prose in the chat | Skill stubs the agent expects |
| Who runs it | The chat host | You, locally |
If you only need a one-off explanation of a file, stay in chat. If you want the agent to reuse the same boundaries on that codebase, you need a skill. The $29 zip scaffolds the stub. It does not finish SKILL.md.
Step 1 — Free local Python preflight (fit check only)
Start with a git checkout you are allowed to read.
Run the free preflight:
Hard boundaries for this step (read them twice):
- Local Python — it inspects license signals and boundary cues on your machine.
- Does not clone — point it at a checkout you already have; it will not fetch a remote for you.
- Does not generate a skill — fit check only; no Agent Skill package from the free step.
- Not legal advice — signals and boundaries for a fit decision; it does not replace counsel.
If the preflight says the repo is out of bounds, stop. Do not pay anyone to paper over a bad fit. This is not a license to distill GPL, AGPL, or LGPL into something you sell.
Step 2 — Scaffold stubs (only after preflight)
When the repo looks like a candidate, Local Repo to Agent Skill Stubs ($29 one-time) gives you one file, SKU-03-skill-distiller-engine.zip. The Python 3 CLI audits the license, scaffolds a skill folder, and checks frontmatter. It does not clone GitHub and does not write the workflow. SKILL.md stays TODO.
- Skill stubs, not a finished skill
- Frontmatter check
- Local CLI you run yourself, not a hosted agent
Shorter workflow overview: Turn a local repo into an Agent Skill
Product page: Local Repo to Agent Skill Stubs
The pack still does not host the agent, open PRs, or invent a finished workflow from a public URL you are not authorized to use.
Step 3 — Install and run locally
Fill SKILL.md, then load the skill in Claude Code, Cursor, or a similar coding agent on your machine. Marketplace discovery and listing rules belong to those tools; this workflow scaffolds a local stub—it does not claim an official Anthropic or Cursor partnership.
You run it. The pack does not babysit the agent for you.
Who this is for
- Builders wiring Claude Code, Cursor, or Windsurf into an existing codebase
- Developers who want checkable stubs from a local repo they can read
- People who will fill SKILL.md and rerun that skill, not ask one chat question
Who should skip this
- Anyone hunting for meeting slides / PowerPoint packs after a call (wrong product line)
- Buyers who want a hosted chatbot with no local project
- People who cannot run a local Python tool
- Anyone who expects the free preflight to clone remotes or emit a finished skill
- Anyone who expects the $29 zip to finish SKILL.md
- Anyone who needs formal legal review of licensing
Suggested order (do not skip the free step)
- Local authorized checkout
- Free preflight (repo-distillation-preflight)
- Read license signals / boundaries — stop if out of bounds
- Local Repo to Agent Skill Stubs, $29 for stubs. SKILL.md stays TODO
- Fill the stub, install in Claude Code / Cursor, and run it yourself
Try the free preflight before you buy. If the fit check fails, you saved $29 and a week of thrash.
FAQ
Does the free tool build the skill?
No. Local Python preflight only. No clone, no skill output, not legal advice.
What does $29 buy?
One zip. The CLI scaffolds skill stubs and checks frontmatter. SKILL.md stays TODO. Not a finished skill, not a hosted agent.
Will this edit my repo?
No. It reads a local path and writes a skill folder. No clone, no PRs, no app changes.
Claude Code and Cursor?
Documented local install paths after you fill the stub. The pack does not claim those runtimes were independently verified.
Workflow and shop links above are the canonical product pages. This Dev.to post is an offsite how-to, not a new thin page on the product site.
Top comments (0)