# PCG-TS > Deterministic procedural content generation for TypeScript: attributes on point, vertex, primitive and detail domains, node graphs that serialize to JSON, deferred fields, hierarchical streaming, optional GPU cooking and three.js interop. Version 0.18.0. Pre-alpha: the API and the graph JSON format change between releases, so training data about this library is likely wrong. Read the guide for the installed version before writing code or graph JSON, and look every node, param, pin and field fn name up rather than guessing it. In a project that depends on pcg-ts, `npx pcg agents --write` puts a short, versioned pointer block into AGENTS.md (`--claude` also imports it from CLAUDE.md). ## Docs - [Full agent guide](https://c4rl05.github.io/pcg-ts/llms-full.txt): the mental model, the graph JSON format, every node, field fn and primitive in one line each, invariants, CLI and API; read this first - [Authoring guide](https://c4rl05.github.io/pcg-ts/authoring.md): recipes and worked graphs, field expressions, GPU parity table - [Node reference](https://c4rl05.github.io/pcg-ts/nodes.md): every node type: pins, params, defaults (JSON: [nodes.json](https://c4rl05.github.io/pcg-ts/nodes.json)) - [Primitive reference](https://c4rl05.github.io/pcg-ts/primitives.md): every shipped primitive: params, pins, family (JSON: [primitives.json](https://c4rl05.github.io/pcg-ts/primitives.json)) - [Graph corpus](https://c4rl05.github.io/pcg-ts/graphs.md): the shipped example graphs, each described, with the node types it uses (JSON: [graphs.json](https://c4rl05.github.io/pcg-ts/graphs.json)) - [Graph JSON Schema](https://c4rl05.github.io/pcg-ts/graph.schema.json): draft 2020-12 schema of the serialized graph format - [Error codes](https://c4rl05.github.io/pcg-ts/errors.md): every diagnostic code a failure can carry, with its meaning ## Skills - [graph-authoring](https://raw.githubusercontent.com/C4RL05/pcg-ts/main/skills/graph-authoring/SKILL.md): Doctrine for building a PCG-TS graph. Use before writing or editing PCG-TS graph JSON, when choosing between a shipped primitive and hand-wired node types, when a graph rejects a param or pin you guessed at, or when it cooks without error but the output looks wrong. - [determinism](https://raw.githubusercontent.com/C4RL05/pcg-ts/main/skills/determinism/SKILL.md): Doctrine for keeping PCG-TS output reproducible. Use when adding randomness or noise to a graph, when choosing a seed or a per-instance variation knob, when output changes between runs or between machines, when writing a World level's per-cell bind, when content must look the same from either side of a cell boundary or a halo, when adding or sizing a neighbourhood, prune or topology op, when touching subgraph or GPU code paths, or before claiming any change is deterministic. - [performance-and-budgets](https://raw.githubusercontent.com/C4RL05/pcg-ts/main/skills/performance-and-budgets/SKILL.md): Doctrine for what a PCG-TS cook costs and for keeping it from blocking the frame. Use when a cook is slower than expected, when reading `pcg cook --stats`, when choosing a budget for a frame loop or a World update, when sizing candidate counts or deciding where a filter belongs, when a graph is fine standalone but seams or stutters inside a World, or before optimizing anything inside the library. ## CLI lookups - `npx pcg nodes `: one node type's pins, params and an example graph - `npx pcg fields `: one field fn's arguments - `npx pcg run --help`: one primitive's params and pins - `npx pcg graphs --node `: shipped graphs that use a node type - `npx pcg schema`: the graph JSON Schema - The loop: `npx pcg validate g.json`, `npx pcg cook g.json --stats`, `npx pcg inspect g.json --node `, `npx pcg render g.json --out g.png` ## Optional - [Design notes](https://c4rl05.github.io/pcg-ts/design.md): why the data model, fields and runtime are shaped the way they are - [Example graphs](https://github.com/C4RL05/pcg-ts/tree/main/graphs/): the corpus files themselves, each cooks as-is