pike/glyph Public Public: anyone can read this; nobody can change it.
Type a prompt, get SVG logos and icons drawn by Claude; preview in your theme, copy or download.
forked from apps/glyph
This repository has no app yet.
Everything here is up to date with apps/glyph.
- contract
- design
- scripts
- src
- static
- .gitignore 37 B · ~11 tok
- AGENTS.md 7.4 KB · ~1.9k tok
- contract.json 766 B · ~256 tok
- forge.json 185 B · ~62 tok
- package.json 391 B · ~131 tok
- README.md 9.4 KB · ~2.4k tok
README.md
glyph
Type a prompt, get SVG logos and icons drawn by Claude; preview in your theme, copy or download.
Describe something — "a hex chip with a circuit b", "a lighthouse for a monitoring app" — pick a kind (logo mark, app icon, badge, spot illustration), and Claude draws four different takes as SVG in about a minute. Switch palette (Sage, Moss, Neon, Mono, Blueprint) to see them in any theme, then Copy the SVG or Save it as a file in that palette.
What is real
- The drawing is real: the app runs Claude Code (
claude -p, no tools,
model claude-opus-5-5) with the token in CLAUDE_CODE_OAUTH_TOKEN. Set it
in Settings → App → Environment (make one with claude setup-token).
Without it a drawing fails at once with a sentence saying exactly that.
- Every SVG is rebuilt from an allowlist (
src/glyph.tssanitize):
shapes, local gradients and clip paths, title and plain attributes
only. No script, style, link, image, url(…) other than one pointing at
a gradient or clip inside the same SVG (url(#id)), or event handler
survives, and previews are <img> data URIs, so
nothing in a drawing can run on the page. A saved file is served with a
sandboxing CSP.
- Colours: Claude draws in three placeholder colours (ink, glow and
paper, design/palettes.json) and builds depth from gradients and
opacity between them; a palette swaps ink and glow for its own and paper
for its ground. Downloads carry the palette's real colours. Drawings are
asked for on a 256-unit canvas with layered, detailed shapes, so a
drawing takes a little longer than a flat one.
- Drawings are private to whoever asked (
contract.json); an admin sees
everyone's. One drawing at a time per person, three at once for everyone. A drawing left half done by a restart is marked failed with a reason.
- Also an API (
/api/drawings,POSTanswers202then the row fills in;
listen on /events for drawing.changed), MCP tools (draw,
get_drawing, list_drawings, delete_drawing) and a terminal (list and
ask; a terminal can't show pictures).
- On the forge an app reaches the internet only through its egress proxy:
forge.json declares api.anthropic.com and claude.ai.
This is an app: the forge runs it. Production follows the default branch;
every open pull gets a preview at its own address with its own copy of the
data. Everything the app keeps lives under FORGE_APP_DATA, which is what
makes "production data" and "preview data" two different things.
PORT=8080 FORGE_APP_DATA=./data bun run src/server.ts # run it here
bun test # what `forge.json`'s app.test runs on every push
FORGE_APP_DATA=./data bun run src/tty.ts # the terminal
FORGE_APP_DATA=./data bun run src/db.ts --check # schema version, integrity
bun run lint:design # no colour but the tokens'
bun run contract:emit # rewrite contract/openapi.json from the routes
app.test in forge.json runs beside app.run in every deploy, not before
it: the preview comes up the moment the app answers, and says live, or
serving · tests failed the moment the suite disagrees.
The shape
One implementation, many doors. src/actions.ts is every operation as a
plain function — (db, who, input) in, data or a named failure out — and
the page, the JSON API, the MCP server and the terminal all call it and
decide nothing themselves. AGENTS.md says how to add to it.
| file | what it is |
|---|---|
src/actions.ts | start, list, read, delete a drawing; the audit trail, read |
src/glyph.ts | the ask, the reply read back, the SVG allowlist, palettes |
src/draw.ts, src/claude.ts | the work after the response: Claude Code, run with no tools |
design/palettes.json | the drawing palettes and the three placeholder colours |
contract.json | who may do what: roles, and the least role each verb needs |
src/contract.ts, src/can.ts | the table, read at boot; the forge's evaluator, verbatim |
src/security.ts | the door: rationing, same origin, the table before the body |
src/audit.ts | every change and every refused one, as a row |
src/who.ts | who is asking, from the forge's signed headers |
src/db.ts | SQLite under FORGE_APP_DATA, WAL, forward-only migrations |
src/live.ts | the in-process bus and GET /events |
src/mcp.ts | the actions as MCP tools at POST /mcp |
src/routes.ts | every path, in one table, each API route described beside it |
src/openapi.ts | the routes as OpenAPI; committed at contract/openapi.json |
src/page.ts | the HTML, rendered on the server, with its share metadata |
design/tokens.json, src/tokens.ts | the look as data, light and dark; compiled to /tokens.css |
src/http.ts | the floor: headers, body cap, files |
src/server.ts | boot: open the database, serve |
src/tty.ts | the terminal, over the same actions |
src/env.ts | what the forge told this process |
static/app.css, static/live.js | the look in the tokens' words, and the board that stays current |
static/glyph.js | copy, the palette swapped in place, the example prompts |
What every app stands on
- The door. Before a route sees a request,
src/security.tsrations
writes (and a stranger's API calls), refuses a write from another site, and turns away — before the body is read — anyone the table lets through a resource's door for no row at all: 401 for a stranger, 403 for anyone else. The action still decides about the real row.
- The audit trail. Every change, and every change refused, is a row:
who, which resource and verb, which row, when, and whether they were
let. Admins read it at GET /api/audit. A row that cannot be written
is a log line, never a failed request.
- The API, described.
GET /openapi.jsoniscontract/openapi.json,
written from the routes by bun run contract:emit; each operation names
its cell of the table (x-contract). bun test fails while it is stale.
- Up, or not.
GET /healthzanswers{ ok, version, uptime }; the
forge waits on it before it sends anyone to a deploy.
- The look.
design/tokens.jsonholds every colour, space and size,
light and dark; nothing else spells out a colour (bun run lint:design),
and every pair a person reads meets 4.5:1 in both themes.
- The open web. A shared link unfurls (description, Open Graph, a
card); /robots.txt keeps crawlers to the page, and out of previews
entirely; /.well-known/security.txt sends a reporter to the forge
that runs the app.
Who you are
The forge says who is asking on every request it proxies, and the app
verifies the signature (src/who.ts); nothing else is trusted. Three tiers:
- anon — anyone. Sees the page and is asked to sign in; nothing else.
- user — signed in to the forge and able to see this app. Draws, as
themselves, and sees and removes their own drawings.
- admin — able to write this repository. Sees and removes any drawing.
"Sign in" is /__forge/signin?to=<path> and "Sign out" is
/__forge/signout; the forge answers both on the app's own host, and
/__forge/report too: it sends anyone, with no account, to report a
problem on this app's repository, if that repository is public.
Any interface
The same actions, the same rules, at every door. A script is the person whose forge grant it carries; a bare script is anonymous.
curl -X POST https://<app>/api/drawings -H 'content-type: application/json' -d '{"prompt":"a fox made of circuit traces","kind":"mark"}'
curl https://<app>/api/drawings/1 # status: drawing → done, with svgs[]
curl -X DELETE https://<app>/api/drawings/1
As an MCP server, for Claude Code or Claude Desktop — draw,
get_drawing, list_drawings, delete_drawing:
claude mcp add --transport http glyph https://<app>/mcp
{
"mcpServers": { "glyph": { "type": "http", "url": "https://<app>/mcp" } }
}
The page stays current over /events: a drawing fills in on its own when
Claude is done, without a reload. Without JavaScript it is the same page, one
reload behind.
The terminal opens with the exact command in FORGE_APP_SSH when the app
runs (ssh -p 2222 <name>@<forge> — the app is the login name, the forge
the host; <name>--<pull>@<forge> for a preview's). It is src/tty.ts on
a real pty over the same data, as the account at the keyboard.
The forge sets PORT, FORGE_APP_DATA, FORGE_APP_ENV (production or
preview), FORGE_APP_URL, FORGE_APP_COMMIT, FORGE_APP_SSH and
FORGE_APP_IDENTITY_KEY when it starts the app. forge.json says how.