What does this skill do, and when should you use it?
canvas-design is one of the 15 skills bundled in the Mini-Agent repository (MiniMax-AI/Mini-Agent, MIT licensed), located at mini_agent/skills/canvas-design/SKILL.md. It targets static visual creation such as posters, art pieces, and graphic designs. The workflow has two steps: author a named design philosophy as a .md file, then express it as a highly visual, text-minimal single-page .pdf or .png. The skill emphasizes minimal text, spatial expression, and expert-level craft, and includes a mandatory second-pass refinement step.
It reads the user's theme or subtle instructions and first outputs a 4-6 paragraph design philosophy .md file (naming an aesthetic movement and articulating space, color, and composition principles). It then deduces a subtle conceptual reference from the original request to serve as the work's 'conceptual DNA,' and produces a single design-forward .pdf or .png page (expandable to multiple pages on request) alongside the philosophy file. It selects fonts from the repository's ./canvas-fonts directory, enforces that nothing overlaps or falls off the canvas, and performs a purification-focused second pass.
- A designer needing a one-page poster for an event, venue, or brand who wants a considered aesthetic stance rather than a template
- A content creator generating a conceptual poster or cover sketch for a book, film, or album
- A design team that wants a quick design-philosophy document plus visual draft as a creative direction starting point
- A Mini-Agent user who wants the agent to handle end-to-end visual production from abstract idea to finished image
- Users needing multi-page visual material, who can request multiple pages bundled into one .pdf forming a tasteful visual narrative
How do you install this skill?
- The skill instructs the agent to download fonts as needed without source constraints; verify that font fetching is constrained by the host environment before use.
- LICENSE.txt and the ./canvas-fonts directory referenced by SKILL.md are absent from the reviewed evidence; license terms and resource availability should be verified independently.
- The skill appears derived from Anthropic's official skills repository without explicit attribution.
- Static review executed nothing; output-quality claims (e.g., 'museum quality') are independently unverified.
- No Chinese-language support statement; the skill assumes the host agent can write files and generate canvas/PDF/PNG artifacts, and may be unusable otherwise.
- Shell / CLI
- Network access
- Local filesystem
canvas-fonts directory for typography (bundled in repository)
The skill ships with the Mini-Agent repository. Quick route: uv tool install git+https://github.com/MiniMax-AI/Mini-Agent.git, then run the official setup script and fill in your MiniMax API key and API base in ~/.mini-agent/config/config.yaml. Development route: git clone the repository, run uv sync, and run git submodule update --init --recursive to initialize the Claude Skills (the skills arrive via submodules). The skill itself lives at mini_agent/skills/canvas-design/. Note: steps for installing it as a standalone Claude Code skill are not documented in the source.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/MiniMax-AI/Mini-Agent.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/mini_agent/skills/canvas-design" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Design a poster for a jazz bar.
Run via the Mini-Agent CLI (mini-agent or mini-agent --workspace /path/to/project) and trigger it in natural language, e.g. 'Design a poster for a jazz bar.' The agent first produces the design philosophy .md, then generates a .pdf or .png. For multi-page work, state the page count in your request. The SKILL.md itself defines no standalone command or trigger syntax; loading depends on the host client.
What are this skill's strengths and limitations?
- Two-phase flow (philosophy first, visual expression second) yields results with a clear aesthetic stance rather than random decoration
- A mandatory second-pass refinement step encodes concrete craft standards: no overlaps, no off-canvas elements, proper margins and breathing room
- A bundled canvas-fonts directory and several worked philosophy examples reduce uncertainty in first-run output quality
- Explicitly requires original design and avoiding copying existing artists' work, lowering copyright risk
- Output quality depends heavily on the model's execution of abstract aesthetic instructions; the source offers no benchmarks or evaluated samples
- Font usage relies on the in-repo ./canvas-fonts directory, making this dependency unclear when installing the skill standalone
- Minimal text is a design goal — it is a poor fit if you need information-dense, text-heavy layout documents
- The specific PDF/PNG generation toolchain (which drawing library, etc.) is not stated in SKILL.md, leaving little basis for troubleshooting
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Canvas Design Skill this page | 42 · Not recommended | ★ 3.1k | 8mo ago | MIT |
| TypeUI Design Fundamentals | 70 · Strongly recommended | ★ 2k | 3mo ago | NOASSERTION |
| Distinctive Frontend Design ✓ Anthropic · Official | 60 · Recommended | ★ 38k | 11d ago | Apache-2.0 |
| Superdesign Design Skill | 53 · Use with care | ★ 840 | 2mo ago | MIT |
| GPT Taste Skill (gpt-taste) | 48 · Use with care | ★ 96k | 11d ago | MIT |
How did FollowSkills review this skill?
Pure prompt file with no code execution, network calls, or sensitive-data handling; low red-line risk. However, SKILL.md instructs the agent to 'download whatever fonts are needed' with no source constraints, an uncontrolled external-fetch instruction; the referenced LICENSE.txt is absent from evidence; no rollback, confirmation, or data-flow disclosure, deducted accordingly.
The skill is self-consistent stepwise instruction (philosophy → canvas → refinement) with no script dependencies to break; but it references a ./canvas-fonts directory not present in evidence, with no failure handling if fonts are missing; no tests or error-feedback design for this skill path; output quality depends entirely on the host agent, deducted accordingly.
Description clearly scopes the use case (posters, art, static design) and output formats (.png/.pdf/.md), giving fairly precise triggers; but non-fit boundaries are undeclared, no Chinese-support statement, and the actual rendering mechanism is unspecified, deducted accordingly.
Frontmatter includes name, description, and license; well-structured, readable, with concrete examples; but the referenced LICENSE.txt is missing from evidence, there is no versioning or changelog, the skill appears derived from Anthropic's official skills repo (README mentions a Claude Skills submodule) without attribution, and maintenance/update responsibility is unclear, deducted accordingly.
Core value is creative design guidance that can steer production of a philosophy .md plus visual artifacts, but static review cannot verify outputs are directly usable; no representative output evidence and no comparative-benefit argument against alternatives, deducted accordingly.
Only the SKILL.md source text is auditable; no execution evidence, no test coverage of this skill path (tests/ cover only agent and ACP), no third-party corroboration; claims like 'museum quality' are unverifiable, deducted accordingly.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →