← ClaudeAtlas

graph-figurelisted

Use when a report, document, or artifact needs a knowledge-graph figure — a picture of entities/concepts/modules and their relationships where the STRUCTURE of connections is the message (dependency maps, concept maps, community clusters). Offers three tiers trading time for quality - full corpus analysis, rendering your own nodes/edges JSON, or a hand-authored SVG. Not for sequences, flows, or quantitative charts.
torifo/skills-graph-figure · ★ 0 · Data & Documents · score 52
Install: claude install-skill torifo/skills-graph-figure
# Graph Figure — knowledge-graph figures in three tiers Produce an embeddable **inline SVG** of a node-link graph for a report. The three tiers trade **time and automation for evidence quality** — pick one explicitly (or ask the user which they want) before doing any work: | Tier | What happens | Time | Quality | Requires | |------|--------------|------|---------|----------| | **a: full** | Analyze a real corpus/codebase with the [graphify](https://github.com/safishamsi/graphify) CLI (AST parsing, community detection), then prune to the relevant subgraph and render | minutes + LLM tokens | highest — extracted from real data | `uv tool install graphifyy` (or `pipx install graphifyy`) | | **b: render** | You (or the agent) curate a `nodes/edges` JSON by hand or from any source, and the bundled renderer lays it out | seconds | depends on your curation | nothing (Python 3, stdlib only) | | **c: hand** | Hand-author a small SVG following the guidance below | minutes of authoring | depends on your drawing | nothing | All three tiers converge on the same deliverable: **one pruned, readable SVG** (top-N nodes or one community — never a full graph; a report figure that needs a magnifying glass has failed). ## Tier selection - Corpus/codebase exists and fidelity matters, and the CLI is installed (or installable) → **a** - You already know the entities and relationships (from reading, from a review, from the user) → **b** (the default workhorse) - Fewer than ~8 nodes, or no Python