A declarative layer where graph affordances are expressed once and mapped — in many ways, against many targets — to UIs, storage, and graph databases. The graph specialization of
zodal.
This repository holds the research, design, and development-planning phase for zodal-graph — there is no package code yet. The research chose and designed the modern, well-maintained tooling for a Zod-v4 schema-driven, renderer-agnostic graph-UI facade, with a strong bias toward reusing existing libraries rather than building from scratch. The planning layer turns that research into an executable, AI-agent-driven build plan.
Design intent
docs/zodal-graph-concept.md— the concept: what zodal-graph is and its three-layer model (Model → Affordances → Targets).docs/graph-affordances-analysis.md— the affordance analysis across twelve graph/timeline subjects ("File 1").docs/graph-zodal-deep-research-prompts.md— the six deep-research prompts (P1–P6).docs/research/— the research reports and decisions.
Planning & toolkit
docs/research_guide.md— routing index: when to read which research doc (doc → purpose → trigger).docs/dev-plan.md— the phased, horizon-graded development plan (living)..claude/CLAUDE.md+skills/— the agent dev guide and thezodal-graphs-dev-*skills that drive the build.
docs/research/README.md is the entry point: the file-naming convention, a status table, and the consolidated tool-decision table (the "money summary"). To find the right deep doc for a task, use docs/research_guide.md; to see what gets built next, read docs/dev-plan.md. For the merge rationale and conflict resolutions, see docs/research/_reconciliation.md; for shared grounding, docs/research/_grounding-brief.md.
Reports are named zgraph_NN<a|b> — a = Claude AI deep-research survey, b = Claude Code grounded report. P1/P2 are b-only; P3 is a-only (grounded during reconciliation); P4/P5/P6 have both.
See the issues for the design/build backlog and discussions for decision records.