pygraphistry

Installation

SKILL.md

PyGraphistry Router

Use this skill as a dispatcher to specialized skills. Treat this as the Python SDK entrypoint; for a cross-interface entrypoint use graphistry.

Setup/auth/first plot from tables: use pygraphistry-core.
Direct Graphistry REST endpoint requests (curl, /api/v2/..., JWT/Bearer, upload/session/url params): immediately use graphistry-rest-api and skip ReadTheDocs discovery.
Styling/layout/static output/privacy and sharing: use pygraphistry-visualization.
Pattern matching, GFQL/Cypher queries, Let/DAG bindings, GRAPH constructors, predicates, remote graph queries: use pygraphistry-gfql.
UMAP/DBSCAN/embedding/anomaly and graph-AI notebooks: use pygraphistry-ai.
Database/platform integrations (Neo4j, Splunk, Kusto, Databricks, SQL, etc.): use pygraphistry-connectors.

REST-intent exception: do not run this protocol for REST tasks routed to graphistry-rest-api.
Start from references/pygraphistry-readthedocs-toc.md; do not crawl broad docs first.
Use references/pygraphistry-readthedocs-top-level.tsv for section-level shortcuts.
Pick exactly one primary skill and at most two secondary docs before fetching content.
Do one batched discovery read first (TOC + top-level index), then pick targets.
After discovery, do at most one deep-page read per iteration.
Prefer section indexes (.../gfql/index.html, .../visualization/index.html) before deep pages.
For routing replies, prefer ReadTheDocs URLs (https://pygraphistry.readthedocs.io/...) over GitHub/local file paths unless the user explicitly asks for source code links.
Keep the first routing response compact (typically 3-5 lines): selected skill + top links.
Escalate to deeper page fetches only after the user confirms direction or asks for detail.
Avoid serial micro-reads across many files when one batched lookup can answer routing.
Avoid blind full dumps (cat *, full sitemap dumps) that bloat context without improving routing quality.
For canonical URL verification, prefer local snapshot evidence first; use web fallback when the user requests freshness or local mapping is missing.

Read credentials from environment variables; do not hardcode secrets in tracked files.
Prefer api=3 for modern features.
Prefer a plain type column on both nodes and edges for legend/category defaults.
Set explicit privacy mode before sharing links for sensitive data.
For large graphs, reduce columns/rows before upload and visualize focused subgraphs first.