Scaffold Claude logo

Scaffold Claude

Community
ragnar-pwninskjold
scaffold-claude

Use when the user wants to create, draft, or scaffold a CLAUDE.md (or AGENTS.md) file for their project. Triggers on "scaffold CLAUDE.md", "write a CLAUDE.md", "set up CLAUDE.md", "/scaffold-claude", or any request to bootstrap project-level Claude/agent instructions. Produces a scratchpad draft that captures edge cases and tribal knowledge — not derivable facts.

Overview

Publisherragnar-pwninskjold
Repositorytech-snacks
Skill namescaffold-claude
Stars
138
Forks
30
Bundled files
3
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 3 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by ragnar-pwninskjold on GitHub. Read the source before you install it.

Installation

Install the Scaffold Claude AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/ragnar-pwninskjold/tech-snacks/tree/main/plugins/tech-snacks/skills/scaffold-claude
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/ragnar-pwninskjold/tech-snacks.git /tmp/tech-snacks
mkdir -p .claude/skills
cp -r /tmp/tech-snacks/plugins/tech-snacks/skills/scaffold-claude .claude/skills/scaffold-claude
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Scaffold Claude in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Scaffold Claude on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Scaffold Claude is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

scaffold-claude

Scaffold a CLAUDE.md for the user's project. Output is written to a scratchpad so the user can review and move it to the repo root themselves.

The file you produce documents edge cases, non-obvious choices, and tribal knowledge — things Claude can't read off package.json, the directory tree, or the README. If a fact is derivable, it does not belong in the file.

The Iron Rule

Never write a section based on inference. Always ask, and stub when the answer is "skip" or "I don't know."

A stubbed section (a <!-- TODO: ... --> comment) is strictly better than a section filled with plausible-sounding guesses. Guesses encode false confidence that future Claude will trust.

When to Use

  • User says /scaffold-claude or asks to "scaffold/write/create CLAUDE.md"
  • User asks for an AGENTS.md (same shape, different filename — ask which)
  • User has a project but no CLAUDE.md and wants one
  • User has a CLAUDE.md they want to redo from scratch

When NOT to Use

  • User wants to edit an existing CLAUDE.md surgically → just edit it, don't run the full interview
  • User wants a tiered, full-codebase AGENTS.md across many directories → use the intent-layer-capture skill instead
  • User wants project-level CLAUDE.md updates triggered by code changes → that's a hook, not this skill

Output Location

Write to: docs/scratchpad/CLAUDE.md (relative to the user's current working directory)

  • Confirm the path with the user before writing.
  • If docs/scratchpad/ doesn't exist, create it.
  • If a file already exists at that path, ask before overwriting.
  • Do NOT write to the repo root. The user moves the file when they're satisfied.

The Workflow

1. Scan (read-only, narrow)

Read these if they exist, only to identify what to ask about — not to fill in:

  • package.json, Gemfile, pyproject.toml, go.mod, Cargo.toml (any one that's present)
  • Top-level directory listing (just names, not contents)
  • README.md (skim, do not summarize into the CLAUDE.md)
  • Existing CLAUDE.md, AGENTS.md, .cursorrules, .github/copilot-instructions.md — if any exist, mention them and ask whether to merge or start fresh

State what you found in one or two sentences. Example: "Saw a Next.js + Drizzle setup with a lib/ directory and no existing CLAUDE.md. I won't assume anything about your conventions — I'll ask."

2. Interview

Walk through the 8 sections one at a time, in order, using the scripts in references/interview-questions.md:

  1. Header paragraph
  2. Stack
  3. Commands
  4. Architecture (only what isn't obvious)
  5. Conventions
  6. Hard constraints
  7. Pointers to deeper docs
  8. Gotchas (tribal knowledge)

For each section:

  • Ask the question(s) from the reference.
  • Always offer "skip" as a valid answer.
  • If the user gives content, echo your paraphrase back before recording it.
  • If the user says "skip" or "I don't know" or "nothing interesting there," stub the section — do not pad it.

Do not batch all questions into one big message. The user's attention is the bottleneck; one section per turn.

3. Write

Use templates/claude-md-stub.md as the base. Fill in confirmed content. Leave every skipped section as the original <!-- TODO: ... --> stub from the template — do not delete the stubs, they help future-you know what was deliberately left blank.

Write to docs/scratchpad/CLAUDE.md after confirming the path.

4. Close

Tell the user:

  • The path you wrote
  • Which sections are filled vs stubbed
  • "Move it to the repo root when you're happy with it. Claude Code auto-loads CLAUDE.md at the root of any project."

No marketing flourishes. No "Now your project is well-documented!" Just the facts.

The One-Shot Example

references/postlane-example.md contains a complete, real-world CLAUDE.md.

Use it to calibrate:

  • The tone (terse, opinionated, every assertion has a because)
  • The shape of each section (what kinds of facts go where)
  • The discipline (nothing derivable from the codebase appears in the file)

Do NOT use it to:

  • Pre-fill the user's file with Postlane's choices
  • Copy any specific tech, command, gotcha, or PR reference
  • Suggest the user "should" have any specific section filled — they shouldn't if they have nothing real to put there

When showing the example to the user, frame it as: "This is what a fully-filled-in version looks like. Yours will be shorter — most projects don't have this much tribal knowledge to document yet, and that's correct."

Rationalizations to Resist

RationalizationReality
"Their package.json clearly shows Next.js — I'll just write that."If Next.js choice has no story, it doesn't belong in Stack. Ask first.
"Most projects use Prettier, I'll add a convention for it."Most projects ≠ this project. Ask.
"They didn't answer for Hard Constraints, but every project has some — let me suggest a few."An empty stub is correct. Invented constraints are worse than no constraints.
"The Gotchas section feels empty, let me add some 'common Next.js gotchas.'"Gotchas are bug-fix archaeology specific to this project. Generic gotchas are noise.
"I'll write 'good code quality' under Conventions as a placeholder."Vague conventions are worse than no conventions. Stub it.
"They said skip but I can tell from the repo that X is true."Skip means skip. The user knows their project; if they skipped, it's because there's nothing non-obvious to say.
"The example has 8 filled sections, theirs should too."The example is for a mature, scarred-by-production codebase. A new project might fill 2 sections. That's correct.

Red Flags — Stop and Recheck

If you find yourself:

  • Reading the codebase to infer content for the file (vs. to identify questions)
  • Suggesting wording before asking the question
  • Filling a section because it "looks empty"
  • Copying any specific Postlane content
  • Telling the user a stubbed section is "incomplete"

Stop. You've drifted into assumption. Re-ask the user; if they say skip, stub it.

Files

  • references/postlane-example.md — the one-shot, with framing for what to copy (structure, voice) vs. not copy (specifics)
  • references/interview-questions.md — section-by-section question scripts
  • templates/claude-md-stub.md — the empty template with TODO comments for each section

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Scaffold Claude AI skill do?

Use when the user wants to create, draft, or scaffold a CLAUDE.md (or AGENTS.md) file for their project. Triggers on "scaffold CLAUDE.md", "write a CLAUDE.md", "set up CLAUDE.md", "/scaffold-claude", or any request to bootstrap project-level Claude/agent instructions. Produces a scratchpad draft that captures edge cases and tribal knowledge — not derivable facts.

Why use Scaffold Claude on TypingMind?

Because you install it once and use it with any model. Scaffold Claude is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Scaffold Claude in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/ragnar-pwninskjold/tech-snacks/tree/main/plugins/tech-snacks/skills/scaffold-claude. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Scaffold Claude?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Scaffold Claude?

As many as you like. As long as a model supports skills, you can use Scaffold Claude with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Scaffold Claude AI skill free?

Yes. It is published on GitHub by ragnar-pwninskjold under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇