scaffold-claude-skill
kevwillow/scaffold-claude-skill
A skill for Claude Code to make Scaffold schema files for whatever you're developing!
Instalação
npx convez-ai kevwillow/scaffold-claude-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
scaffold-gui — a Claude Code skill for Scaffold
Generate Scaffold artifacts from inside a Claude Code session. Scaffold is an offline desktop GUI that renders native forms from JSON "schemas" describing CLI tools, then assembles and runs the command for you.
This repo packages the scaffold-gui skill two ways: as an installable Claude
Code plugin (via a plugin marketplace) and as a manual-clone skill folder
you copy into ~/.claude/skills/.
What the skill does
Two capabilities, each driven by a canonical prompt bundled in the skill:
- Schema generation — turn a CLI tool's documentation (
--helpoutput, a man page, or a docs URL) into a valid Scaffold tool schema (_format: "scaffold_schema"), then auto-validate it before writing<binary>.json. - Preset generation — given an existing tool schema plus a natural-language
description, produce one or more saved form configurations (
_format: "scaffold_preset"), each cross-checked against the schema, then written as<name>.jsonfiles.
The integration value is the workflow around the prompts: applying them verbatim,
auto-validating the output with the bundled stdlib-only validators
(validate_schema.py, validate_preset.py), and placing files where you can
commit them. The skill does not generate cascades, and does not modify
Scaffold's application code.
Bundled files (skills/scaffold-gui/)
| File | Purpose |
|---|---|
SKILL.md | The skill definition and workflow Claude follows. |
SCHEMA_PROMPT.txt | Canonical CLI→schema conversion prompt (applied verbatim). |
PRESET_PROMPT.txt | Canonical preset-generation prompt (applied verbatim). |
schema.md | The Scaffold JSON schema specification (the rules). |
validate_schema.py | Stdlib-only schema validator (python 3, json only). |
validate_preset.py | Stdlib-only preset validator (cross-checks against a schema). |
Mirrored content.
SCHEMA_PROMPT.txt,PRESET_PROMPT.txt, andschema.mdare mirrored byte-for-byte from the canonical Scaffold repository, which is the source of truth. If they ever diverge, the Scaffold repo wins — refresh the copies here from there.
Target schema-format version
The bundled prompts and validators target Scaffold schema specification
Version 1.0 (see schema.md), i.e. the scaffold_schema / scaffold_preset
formats.
Prerequisites
- Claude Code (recent version with skills/plugins support).
- Python 3.10+ on your
PATH— the validators run with the standard library only (nopip install, no third-party packages, no network access). - To actually use the generated files, the Scaffold desktop app. Generating and validating schemas/presets here needs only Claude Code + Python.
Install
Option A — Plugin marketplace (recommended)
This repo is itself a Claude Code plugin marketplace containing one plugin
(scaffold-gui). Inside Claude Code:
/plugin marketplace add kevwillow/scaffold-claude-skill
/plugin install scaffold-gui@scaffold-claude-skill
To update later, after new commits are pushed:
/plugin marketplace update scaffold-claude-skill
When the plugin is enabled, Claude Code discovers the skill at
skills/scaffold-gui/SKILL.md and exposes it (namespaced as
scaffold-gui:scaffold-gui).
Option B — Manual clone into your personal skills
Clone the repo and copy the self-contained skill folder into your personal skills directory.
macOS / Linux:
git clone https://github.com/kevwillow/scaffold-claude-skill.git
cp -r scaffold-claude-skill/skills/scaffold-gui ~/.claude/skills/scaffold-gui
Windows (PowerShell):
git clone https://github.com/kevwillow/scaffold-claude-skill.git
Copy-Item -Recurse scaffold-claude-skill\skills\scaffold-gui $HOME\.claude\skills\scaffold-gui
Claude Code discovers personal skills at ~/.claude/skills/<name>/SKILL.md. If
the ~/.claude/skills/ directory did not exist when Claude Code started, restart
it so the new directory is watched.
Usage
Once installed, ask Claude (or invoke the skill directly). One-liners:
- Schema: "Use the scaffold skill to generate a schema for
ripgrep --help." (Paste the--helpoutput, give a man page, or provide a docs URL.) - Preset: "Use the scaffold skill to make a preset for
tools/ripgrep.json: a case-insensitive recursive search of the current directory." (Preset generation requires an existing schema as input.)
Claude applies the bundled prompt, validates the result, and writes the committable
.json file(s).
Use it on your own CLI (end-to-end)
The full loop for scaffolding a GUI for your own program. Schema and preset generation happen in Claude Code (this skill); the Scaffold desktop app then consumes the JSON and renders the form.
1. Prerequisites
- The Scaffold desktop app installed — it's what turns the JSON into a GUI (see the Scaffold repo).
- This skill installed via either path above (plugin marketplace or manual clone), plus Python 3.10+ for the validators.
Generation, validation, and Scaffold itself are all local — nothing here needs the network.
2. Generate a schema for your tool
Preferred — project-context mode. Run the skill from inside your CLI's
repo and ask it to build the schema. Instead of relying on --help alone, it
reads your argument-parser source (argparse/click/cobra/clap/… or a
hand-rolled parser) plus the README/docs//man pages to recover the rich
relationships --help can't give — mutually-exclusive flag groups,
depends_on dependencies, enum choices, defaults, and real hover
descriptions:
"Use the scaffold skill and my project's context to build a Scaffold schema for
greet."
Fallback — paste mode. If you don't have the repo handy, give it the docs directly:
"Use the scaffold skill to generate a Scaffold schema for
greet --help:" then paste the help text (or give the man page / docs URL).
Either way, the skill applies the canonical SCHEMA_PROMPT.txt verbatim,
validates the result with validate_schema.py, and writes greet.json (named
after the binary).
3. Put greet.json where Scaffold finds it — the key step
Scaffold's tool picker scans a tools/ folder (including subfolders) and lists
every .json schema in it. Drop your greet.json there. Which tools/ folder
depends on how Scaffold runs:
| How Scaffold runs | Put your schema in |
|---|---|
| Installed app — Windows | %APPDATA%\Scaffold\tools\ |
| Installed app — macOS | ~/Library/Application Support/Scaffold/tools/ |
| Installed app — Linux | ~/.local/share/Scaffold/tools/ |
| From source / portable build | the tools/ folder next to scaffold.py |
The bundled schemas in that folder are read-only; your own go in the same writable
user tools/ folder. The filename convention is <binary>.json, but Scaffold
reads the tool/binary fields inside the file, so any .json name works.
Reopen Scaffold (or its tool picker) and greet appears in the list. (Grounded in
scaffold.py's _tools_dir() / ToolPicker.scan() and the "User data" section of
Scaffold's README.)
4. Generate presets for your tool
Presets require an existing schema. Give Claude the schema path plus a plain-English description of what you want:
"Use the scaffold skill to make a preset for
tools/greet.json: shout 'HELLO' three times."
The skill validates each preset against that schema and writes it with a
lowercase-underscore filename (e.g. shout_hello.json). Presets live under the
same user-data root as step 3:
<user-data>/presets/<tool>/<name>.json
where <tool> is the schema's tool field (here greet) — for example
%APPDATA%\Scaffold\presets\greet\shout_hello.json (installed, Windows) or
presets/greet/shout_hello.json next to scaffold.py (source mode). Scaffold's
preset picker lists every .json in that tool's folder. (Grounded in
scaffold.py's _presets_dir().)
5. Order & offline
Generate the schema first, then any presets — a preset is meaningless without its schema. None of these steps — generation, validation, or Scaffold rendering and running the command — requires internet access.
License
This distribution is licensed under the MIT License — see LICENSE.
The bundled prompts (SCHEMA_PROMPT.txt, PRESET_PROMPT.txt) and schema.md are
mirrored byte-for-byte from the canonical Scaffold repository, which remains
the source of truth for that content (see the Mirrored content note above).