hand-drawn-diagrams

Tier-2 Review: hand-drawn-diagrams (Cluster: document-generation)

What we attempted

We ran three standard harness tests against this skill: install, minimal-roundtrip, and edge-corner. The intent was to verify that the skill could (a) be installed in a clean container, (b) take a simple prompt and produce a usable diagram artifact (SVG, PNG, or hosted link), and (c) gracefully handle edge cases like embedded media or complex shapes. The skill’s summary looked strong — clear trigger phrasing, specific output claims (animated SVG, hosted edit link, PNG export), and a claimed compatibility with multiple agent runtimes.

What failed — and why

All three tests failed. The blocker was not the skill’s conceptual design, but a structural omission in SKILL.md: it contains no installation instructions, no dependency list, no operational details (e.g., how to invoke the Excalidraw CLI or any conversion utility), and no edge-case handling notes. Concretely:

• install (fail): The skill says it “works with Claude Code, Codex, Gemini CLI” but never lists a single command, package, or environment variable. We could not install any required tool in a clean container. There is no mention of npm, pip, a binary, or a Docker image. The skill is effectively a description of a behavior, not a runnable unit.
• minimal-roundtrip (fail): The skill describes generating diagrams “from a prompt,” but does not specify whether it reads an input file, transforms it, or writes back to disk. No target format details are given — is the output a single SVG, a zip, a JSON file for Excalidraw, or a URL? Without file I/O semantics, we could not simulate a roundtrip.
• edge-corner (fail): No mention of how to handle embedded media (images, links), complex shapes (arrows, grouping, LaTeX), or conversion limitations (e.g., font fallback, scaling, or lossy PNG export). There is no failure-mode documentation — what happens if the prompt is ambiguous, or if the Excalidraw API is down?

What we observed

The SKILL.md content is essentially a marketing blurb: “Generate hand-drawn Excalidraw diagrams from a prompt — animated SVG, hosted edit link, and PNG export.” It names target agents but provides zero operational scaffolding. The dimension scores (D1–D5) were generated from the text’s apparent clarity, but those scores are theoretical. No test could pass because the skill never bridges from intent to execution. There were no observed dependencies — not because none exist, but because none were declared.

Rating caveat

The composite score of 4.2 / 5.0 is theoretical and provisional. It reflects the quality of the description — the triggers are clear, the output claims are specific, and the scope is well-bounded. However, until the skill is re-run in a physical environment with a complete SKILL.md (installation steps, dependency manifest, file I/O contract, and edge-case policy), the score cannot be validated. We are not downgrading the skill’s potential, but we are flagging that the current artifact is not deployable as-is.

Is the skill valuable in principle?

Yes — in principle, this is a genuinely useful skill. Hand-drawn diagram generation from natural language is a real need for documentation, whiteboarding sessions, and design feedback loops. The promise of an animated SVG plus a hosted edit link is compelling, and the multi-agent compatibility claim suggests broad applicability. The core idea is sound, and the output format (Excalidraw) is a well-known, open standard. The failure is purely one of packaging, not concept. With a modest revision — adding a requirements.txt or package.json, a usage example, and a short “Known Limitations” section — this skill could plausibly score 4.5+ and pass all three tests. We encourage the creator to add operational detail; the underlying value is real.

Final word: This review is honest signal, not a takedown. The skill is promising but currently incomplete. Re-run after the author supplies installation and operational details — we expect it to pass.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-08-30.

Overall: broken. All 3 tests failed; key blocker: SKILL.md lacks installation instructions, operational details, and edge-case handling, making it impossible to simulate realistic test outcomes.

Test	Status	Notes
install	fail	SKILL.md does not provide an install command or dependency list; no way to install required tools (e.g., Excalidraw CLI or conversion utilities) in a clean container.
minimal-roundtrip	fail	SKILL.md describes generating diagrams from prompts but does not specify file reading, transformation, or write-back operations; no target format details provided.
edge-corner	fail	SKILL.md does not mention handling edge cases like embedded media, complex shapes, or conversion limitations; no details on supported features or failure modes.