lsp-max

Law-state LSP runtime that projects a multidimensional state machine through LSP 3.18. Provides maximum protocol coverage, process-mining conformance, and receipt-chain admission for agents, CI systems, and release gates—not tower-lsp, but a distinct runtime with extended max/* surfaces for introspection, gates, snapshots, and law enforcement.

Quick start

Clone and build:

git clone https://github.com/seanchatmangpt/lsp-max.git
cd lsp-max
just setup            # clones the sibling repos this workspace needs to build
cargo test --workspace

This workspace does not build standalone: it depends on three sibling checkouts (../lsp-types-max, ../wasm4pm-compat, ../wasm4pm). just setup (or bash scripts/bootstrap.sh) fetches them; just doctor reports what is missing without changing anything. In Claude Code on the web, a SessionStart hook runs the bootstrap automatically, so sessions start build-ready.

Run an example:

cargo run --example anti-llm-cheat-lsp

Use in your project:

[dependencies]
lsp-max = "26.6"

Key features

Directory structure

lsp-max/
├── src/                    # LSP server framework core
│   ├── lib.rs             # Main crate interface
│   ├── language_server.rs # LanguageServer trait definition
│   ├── service.rs         # LspService orchestration
│   ├── gate.rs            # Λ_CD gate state machine
│   ├── diagnostics.rs     # Law-state diagnostic engine
│   ├── composition/       # Multi-server compositor internals
│   └── primitives/        # Law-state value types (Receipt, ConformanceVector, etc.)
│
├── lsp-max-protocol/      # max/* method declarations, capability vectors, receipts
├── lsp-max-macros/        # Proc macros (#[lsp_max::async_trait], etc.)
├── lsp-max-runtime/       # Typestate machine, phases, transitions, snapshots
├── lsp-max-agent/         # Agent integration, analysis bundles
│
├── crates/
│   ├── lsp-max-cli/       # Noun/verb CLI (clap-noun-verb-based actuation grammar)
│   ├── lsp-max-client/    # LSP client framework (drives servers in tests)
│   ├── lsp-max-compositor/ # Multi-server hub with gate, quorum debounce, receipts
│   ├── lsp-max-base/      # Base LSP type extensions
│   ├── lsp-max-live/      # Live protocol surfaces (streaming diagnostics)
│   ├── lsp-max-lsif/      # LSIF export and conformance checking
│   ├── lsp-max-specgen/   # Codegen from LSP 3.18 metaModel.json
│   ├── lsp-max-adapters/  # Tree-sitter-driven AST/codegen (ported auto-lsp stack)
│   ├── playground/        # Dev-dependency test harness with demo binaries
│   └── lsif-*/            # LSIF indexing and linking tools
│
├── examples/              # Domain-specific dogfood servers
│   ├── anti-llm-cheat-lsp/  # Diagnostic canary (detects tower-lsp, fake receipts, victory language)
│   ├── clap-noun-verb-lsp/  # Noun/verb CLI demo
│   ├── pattern-lsp/         # Pattern detection
│   ├── wasm4pm-lsp/         # Process-mining over wasm4pm
│   ├── axum-lsp/            # Axum framework integration
│   ├── bevy-lsp/            # Bevy framework integration
│   ├── tex-lsp/             # TeX/LaTeX LSP
│   └── *.rs                 # Explanation crates (receipt_chain_explained, conformance_vector_explained, etc.)
│
├── tests/                 # Integration and e2e tests
│   ├── e2e/              # End-to-end test suites
│   ├── lsp318_capabilities/ # LSP 3.18 feature validation
│   ├── dogfood_loop/     # Dogfood test drivers
│   └── common/           # Shared test utilities
│
├── web/                  # Next.js dashboard and analytics
│   ├── app/             # Next.js App Router
│   ├── lib/             # Shared components and utilities
│   └── REPRESENTATION_MAP.md  # Dashboard data model
│
├── docs/                # Narrative and reference documentation
│   ├── FEATURES.md              # LSP 3.18 coverage matrix with receipts
│   ├── EXAMPLES.md              # Diataxis-mapped example index
│   ├── TEST_INFRA.md            # Test architecture and conformance
│   ├── CANCELLATION_SAFETY.md   # Async cancellation guarantees
│   ├── ADR/                     # Architectural Decision Records
│   ├── law/                     # Law-state semantics papers
│   ├── papers/                  # Academic references
│   ├── reports/                 # Analysis and audit reports
│   └── thesis/                  # Theoretical foundation documents
│
├── CLAUDE.md            # Codebase laws and instructions (enforced by anti-llm-cheat-lsp)
├── AGENTS.md            # Agent protocol, conformance laws, admission rules
├── CONTRIBUTING.md      # Development guidelines
├── Justfile             # Build orchestration (just test, just dx-verify, etc.)
└── README.md            # This file

Published crates

All other workspace crates have publish = false.

Design principles

• No victory language: Status values are bounded (ADMITTED, CANDIDATE, BLOCKED, REFUSED, UNKNOWN, PARTIAL, OPEN) — never "done," "solved," or "guaranteed."
• Unknown is preserved: ConformanceVector axes never collapse unknown into admitted or refused; ambiguity is explicit.
• Receipts, not logs: Capability claims require BLAKE3-hashed receipt artifacts with path, digest, boundary, and checkpoint — test stdout is not a receipt.
• Read-only LSP surface: The server emits diagnostics, hovers, and intents but never mutates files; all mutation is client-driven.
• CalVer, not SemVer: Version 26.6.24 encodes the date (2026-06-24); mismatches are diagnostic events.
• Distinct from tower-lsp: Never reference plain tower_lsp or tower-lsp in code, manifests, or docs (outside negative-control fixtures).

Build & test

Run just alone to list recipes:

just test                 # cargo test --workspace
just test-e2e             # cargo test --test e2e
just test-pre-publish     # dx-verify + dx-polish + all tests (≤10s)
just dx-polish            # cargo fmt --all + clippy -D warnings
just dx-verify            # Architectural boundary scan across sibling repos

Single test or crate:

cargo test -p lsp-max
cargo test --test test_lsp318_capabilities
cargo test -p anti-llm-cheat-lsp --test dogfood

Sibling dependencies

The workspace requires sibling checkouts at:

• ../lsp-types-max — LSP 3.18 type authority (with proposed features)
• ../wasm4pm-compat — Process-mining type baseline
• ../wasm4pm — Execution engine

No intermediary type crates (wasm4pm_types, ocel_core, etc.) are allowed.

Documentation

• GETTING_STARTED.md — Setup and first server
• CONTRIBUTING.md — Development workflow and law enforcement
• docs/ARCHITECTURE.md — Five-layer model, typestate machine, Λ_CD gate
• crates/anti-llm-cheat-lsp/docs/COVERAGE_LSP318_LSIF06.md — Authoritative LSP 3.18 (95-method) + LSIF 0.6 (38-element) combinatorial coverage matrix
• docs/FEATURES.md — Legacy per-version feature table (superseded by COVERAGE_LSP318_LSIF06.md for LSP 3.18)
• DOC_COVERAGE_LOG.md — Documentation audit and gaps
• CLAUDE.md — Codebase laws and constraints (consulted by anti-llm-cheat-lsp)
• AGENTS.md — Agent protocol and conformance rules

License

Dual-licensed under MIT or Apache 2.0 at your option.

Rust version: 1.70+ | CalVer: 26.6.24 (2026-06-24) | Status: CANDIDATE