Skip to content
Agentmart
Submit a tool

collage-design

Create beautiful cut-and-layer collage art in .svg and .png using a collage philosophy, sourcing real open-licensed imagery and composing it into a single…

Skill by polgarp

IndexedApache-2.0★ 4
Write a review

Chalk Labs

Need help deploying collage-design?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/polgarp-collage-design

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Create beautiful cut-and-layer collage art in .svg and .png using a collage philosophy, sourcing real open-licensed imagery and composing it into a single crafted piece. Use this skill whenever the user asks for a collage, photomontage, cutout composition, cut-and-paste artwork, mixed-media poster, or any single artwork made by cutting and layering multiple found/photographic images into one composition — even if they don't say the word "collage." This is for creating a finished art piece, not arranging a moodboard grid, removing a background from one image, stitching a panorama, or generating art from code. Sources only open-licensed images and logs attributions; never copies living artists' compositions.

Use cases: documents, design

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Python package

Network hosts it talks to 2

  • codeload.github.com · found in install command
  • github.com · found in code

Files it touches

  • ~/.claude/skills/collage-design
  • /Library/Fonts
  • /opt/homebrew/share/fonts
  • /usr/local/share/fonts
  • /usr/share/fonts
  • /usr/share/fonts/truetype/dejavu/DejaVuSans.ttf
  • Your home directory (~)
  • ~/.fonts
  • ~/.local/share/fonts
  • ~/Library/Fonts

What's inside

Skill name
collage-design
Description
Create beautiful cut-and-layer collage art in .svg and .png using a collage philosophy, sourcing real open-licensed imagery and composing it into a single crafted piece. Use this skill whenever the user asks for a collage, photomontage, cutout composition, cut-and-paste artwork, mixed-media poster, or any single artwork made by cutting and layering multiple found/photographic images into one composition — even if they don't say the word "collage." This is for creating a finished art piece, not arranging a moodboard grid, removing a background from one image, stitching a panorama, or generating art from code. Sources only open-licensed images and logs attributions; never copies living artists' compositions.
License
Apache-2.0
Instructions
526 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Agent Skills spec issues

  • body: 526 lines (spec recommends under 500)

Files (27)

  • marketplace.json · .claude-plugin/
  • plugin.json · .claude-plugin/
  • .gitignore
  • LICENSE.txt
  • NOTICE.md
  • README.md
  • SKILL.md
  • README.md · evals/
  • trigger-evals.json · evals/
  • brine.jpg · examples/
  • mauve.jpg · examples/
  • sodium.jpg · examples/
  • strike-anywhere.jpg · examples/
  • pipeline-recipes.md · references/
  • batch.py · scripts/
  • check_batch.sh · scripts/
  • check_deps.sh · scripts/
  • check_render.sh · scripts/
  • cut.py · scripts/
  • fetch.py · scripts/
  • fonts.py · scripts/
  • knockout.py · scripts/
  • setup_fonts.sh · scripts/
  • survey.py · scripts/
  • svgkit.py · scripts/
  • treat.py · scripts/
  • vectorize_text.sh · scripts/

SKILL.md (first lines)

Instructions for making **collage** — art assembled from real, found imagery: sourced, cut,
layered and reconciled into a single object, with text laid over as the philosophy requires.
Built from **actual photographs and printed material found on the open web**, under
open licences only.

The deliverables are `brief.md`, `philosophy.md`, `attributions.md`, a self-contained `.svg`, its `.png`
render, and the build script that made them.

First **choose the graft** with the user (three proposals, one pick), then four movements:
1. **Collage philosophy** (`philosophy.md`)
2. **Deduce the subtle reference**
3. **Source the material** (open-licensed or user-supplied → `sources/` + `attributions.md`)
4. **Compose on an SVG substrate** → `.svg` + `.png` + build script, then a critique against the brief and a polish pass.

### BEFORE STARTING

- **Save the brief verbatim** to `brief.md`: the user's words and anything they pointed to.
  Below it, sort the request into three short lists: **content** (every word and fact the piece
  must carry: title, date, venue, line-up), **style direction**, and **avoid** (exclusions the
  user stated). Leave a list empty rather than inventing for it. Movement 1 writes the
  philosophy without the commission on purpose; this file is what the critique before the polish
  pass checks the piece against.
- **Real imagery is mandatory** — either downloaded from the open web (needs `WebSearch`/`WebFetch`
  or browser tools) or taken from a folder the user points you at. With neither available, say so
  and stop; never substitute generated or invented imagery.
- **Local tooling:** ImageMagick, Inkscape, librsvg, fontconfig, Python + Pillow + NumPy. Run
  `scripts/check_deps.sh` first; it prints the install line for anything missing.
- **The bundled toolkit is a reference implementation, not the requirement.** `scripts/` covers
  fetching, surveying the pool, edge-cutting, silhouette extraction, unifying treatments and SVG
  assembly; `references/pipeline-recipes.md` holds the exact commands and the traps for each. Use
  them — they encode failures that are expensive to rediscover. But what this skill actually
  demands is the judgement in Movements 1–3 and the invariants in Movement 4. If you have better
  tools, use them and meet the same bar.
- **This is a long job** — if the user wants something quick, tell them before Movement 1. A
  multi-piece **series** costs far less than N× a single piece: sourcing dominates the budget, and
  one source pool feeds every piece in the set. In a series, every piece takes its words from
  `brief.md` alone. A tagline invented for one piece must not quietly turn up in the next.

---

## THE DEFAULT REGISTER — PHYSICALITY OVER DIAGRAM

A collage is **physical**. Fragments overlap, occlude, and touch; cut subjects escape their
frames; torn edges show their fiber; the surface accumulates. Unless the philosophy decides
otherwise *in writing*, the piece must read instantly as **cut paper layered by hand** — not as a
clean diagram with a few small photos inset into tidy, separated boxes.

Left alone, composition slides toward the grid: neat rectangles, one timid image each, evenly
spaced, floating in polite empty space. **If the piece could be rebuilt in PowerPoint with nothing
lost, it drifted.**

**The defaults:**
- Fragments **overlap and occlude** — things sit in front of other things.
- Some subjects are **silhouette knockouts that break their bounding box** — a hand, a figure, a
  specimen spilling past any implied frame. This is settled at *sourcing*, not composition: it
  depends on finding material a subject can be lifted out of, which is Movement 3's job and is
  measured there rather than guessed. Most pools contain some; plates, sequences and anything shot
  against a sweep or a night ground are usually rich in them, and the measurement finds them per
  *crop*, so a source that looks hopeless whole is often not. A pool that genuinely holds none —
  labels and ephemera cropped to their own printed border, say — takes the escape from a fragment
  bleeding off the **canvas edge** instead. Reach for that because the material was measured and
  came back empty, not before looking.
- **Edges are visible and physical** — torn fiber, knife-cut bevel, cast shadow. Every fragment
  shows how it was severed; the edge is where the piece declares it was cut by someone.
- **Density is a virtue** — and it is a matter of *count and scale variance*, not coverage: many
  fragments, the largest several times the smallest. A dozen mid-sized pieces at even spacing is
  the drift however much of the page they cover. Open space only where text must be read.

### OVERRIDING A DEFAULT

Any default above can be set aside — deliberate tonal clash (Höch, Heartfield), the rigorous
photographic grid, the near-empty field holding one fragment, glitch and scan artifact,
type-as-image are all available registers.

Override **in the philosophy, by name, with a reason** — one sentence saying which default is set
aside and what the piece gets for it:

> *"Fragments are NOT tonally unified: each source keeps the colour temperature it arrived with,
> because the collision between registers is the argument the piece is making."*

**Then say what replaces it.** An override removes a constraint; it has to add one, or the piece
reads as a mistake rather than a decision — *two treatments that never cross, curved edges for one
family and straight for the other, a fixed paint order*. What replaces it is a sourcing and
cutting decision, so it belongs here, before Movement 3 spends anything on material.

Named there, the override governs the rest of the run and the Movement 4 self-check validates
against it. Unnamed, the default stands.

---

## CHOOSE THE GRAFT

The aim is a piece that looks made and that nobody has quite seen before. A collage tradition
supplies the first: it comes with a way of cutting, a way of stacking and a craft bar. But a
tradition alone is a pastiche. Novelty comes from **crossing one collage lineage with one
vernacular from outside art**: the lineage decides the cut and the finish, and the vernacular
decides the system, and often supplies the material as well. `references/registers.md` lists both
and explains how to pair them. Read it now.

**1. Name the default.** In your working notes (not a deliverable), describe in a few lines the
collage this brief gets if nobody decides. That covers two things: the genre template for this
kind of commission, and this skill's own pull (sepia archival ageing, the Western institutional
archive, specimen plates and documents, a muted clamped palette). That is the starting point to
move away from. Any part of it can still be chosen later, for a stated reason.

**2. Propose three grafts**, each one lineage × one vernacular. Keep them far from the default and
from each other: no shared parent, and at least one should be loud. Give each in a single line:

> **Décollage tide table** — torn layers of harbour posters over NOAA tide columns; the dates
> live in the table; material: NOAA tables (PD) plus poster photography on Commons.

That is the graft phrase, what the vernacular's system carries from the brief, and where the
material would come from. The last part is what makes this collage-specific. A graft whose
material doesn't exist under an open licence is a dead end, so do a search or two per graft to
check before proposing it rather than after the user has chosen it.

**3. Stop and let the user pick.** They choose one, cross two, or send you back for three more.
If the intended use is still unknown (see licence compatibility, Movement 3), ask it in the same
message, so the user is only stopped once. Add the chosen graft to `brief.md`; it is part of the

From polgarp/collage-design, Apache-2.0.

Signals

Stars
4
Forks
0
Open issues
0
Last commit
Oct 3, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get collage-design running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying collage-design?

CallGet help
Cite this page

collage-design skill. Agent Mart. https://agentmart.dev/skills/polgarp-collage-design (last updated 2026-10-24)