graphistry-skills

Skill files for AI agents (including Claude Code and OpenAI Codex) to better use the Graphistry ecosystem.

Graphistry is a graph intelligence ecosystem with fast-moving capabilities across graph ETL/shaping, visualization, GFQL graph querying, and AI workflows. These skills help agents use more of that surface area correctly and reach good results faster.

Strong frontier models often already know core Graphistry/PyGraphistry patterns due to ecosystem maturity and backward compatibility. The skills add high-value guidance on newer features, preferred workflow patterns, and safer/more reliable execution details.

Install

npx skills add graphistry/graphistry-skills \
  --agent codex \
  --agent claude-code \
  --skill pygraphistry \
  --skill pygraphistry-core \
  --skill pygraphistry-gfql \
  --skill pygraphistry-visualization \
  --skill pygraphistry-ai \
  --skill pygraphistry-connectors \
  --yes

Claude Code Example (Live URL)

Run from a project where these skills are installed and graphistry + pandas are available.

export GRAPHISTRY_USERNAME="your_user"
export GRAPHISTRY_PASSWORD="your_pass"
export GRAPHISTRY_SERVER="hub.graphistry.com"
export GRAPHISTRY_PROTOCOL="https"

PROMPT='Using Bash tool calls, run (without creating files) a tiny PyGraphistry
cyber hunt demo (5-10 rows) with realistic devices/users/processes/ips/domains
and event entities that include explicit event_time timestamps, include node and
edge type fields, style with icons plus risk coloring, set
graphistry.privacy(mode='"'"'public'"'"', notify=False), call plot(render=False),
and print only the final live URL.'

claude -p \
  --model opus \
  --permission-mode bypassPermissions \
  --tools Bash \
  "$PROMPT"

Sample output (validated on 2026-02-21, model=opus, runtime ~68.2s):

https://hub.graphistry.com/graph/graph.html?dataset=17743ba9ff3549729fdb4d9c1c071bbc&type=arrow&viztoken=e968954a-c0e5-4206-85a6-3d950817a6d4&usertag=ef9e6f8d-pygraphistry-0.50.6&splashAfter=1771659185&info=true

Evals

These skills are regularly benchmarked and tuned against standard Graphistry user journeys (baseline vs skills, multiple runtimes/models).

For reproducible commands and sweep workflows, see DEVELOP.md.

Current checked-in benchmark packs show skills improving pass rates significantly:

• Fresh eval sweep with isolated baseline (codex, skills=both, 56 cases × 2):
  • skills=on: 91% pass (51/56), avg 47.4s
  • skills=off: 52% pass (29/56), avg 46.4s
  • Delta: +39pp pass rate improvement
• Prior sweep for reference (note: had baseline contamination bug):
  • skills=on: 88/100 pass
  • skills=off: 81/100 pass

See:

• benchmarks/reports/2026-03-01-baseline-isolation-sweep.md - Latest sweep with baseline fix
• benchmarks/reports/2026-02-23-postcleanup-fullsweep.md - Prior sweep

Docs

• Contributor/dev workflows and sweep commands
• Contributing guide
• Security policy
• Code of conduct
• License
• Report generator
• Scenario coverage audit tool
• Benchmark artifact structure
• Latest eval sweep plan