Documentation

Everything you need to integrate MustardScript into your AI agent stack.

Getting Started

Real-world programmatic tool-calling examples for code mode, multi-tool execution, and resumable workflows

Describes the implemented language surface, supported value types, syntax, and built-in methods

Internal guest-value heap model, GC semantics, and value representation

Stack-based bytecode VM architecture and instruction set

Instruction budgets, heap limits, call depth, and timeout controls

Host boundary, structured values, capability calls, and error handling

Queue-oriented executor API for managing resumable executions

Sidecar process wire protocol using newline-delimited JSON

Compiled program and snapshot serialization formats

Normative security contract and threat assumptions

Project-level threat model, security objectives, and mitigation strategies

Hostile input testing, fuzzing entry points, and hardening strategies

Node-parity conformance buckets and validator-rejected test strategy

Performance metrics, optimization results, and latency measurements

Phase-2 portfolio coverage matrix for the audited programmatic tool-call gallery

Release process, verification checklist, and publishing steps

Audit status of the realistic programmatic tool-call use case gallery

Architecture decision record for Rust core + Node addon + sidecar design