Agent eXecution Model — deterministic tooling and structured Protocols for AI agents.

axm-protocols starts from a simple observation: in agent workflows, not everything needs to be probabilistic. LLM generation is inherently non-deterministic — but code analysis, quality checks, and workflow orchestration can be entirely deterministic. AXM builds that deterministic layer.

• Structured output — every tool returns JSON, designed for agents that consume data, not parse text
• AST-powered analysis — tree-sitter based code intelligence for semantic precision, not grep noise
• Codified quality gates — 40+ rules across lint, types, coverage, complexity, security, and governance
• Reproducible workflows — conventional commits, pre-commit enforcement, semantic versioning

AST introspection, code auditing, project scaffolding, and git automation.

Package	Description	Version
axm-ast	Code intelligence via tree-sitter — callers, impact, dependency graphs
axm-audit	Code quality auditing — 6-category composite scoring on 100 pts
axm-init	Project scaffolding & 39-check governance gate
axm-git	Structured commits, semantic tagging, preflight checks

uv add axm-ast axm-audit axm-init axm-git

Unified CLI, protocol execution engine, and MCP server. Nexus provides the runtime that turns structured YAML protocols into validated, checkpointed agent workflows.

Coming soon — currently in development.

Every AXM package is measured on two axes:

• Governance (axm-init check) — src/ layout, PEP 621, py.typed, GitHub Actions CI, MkDocs docs, pre-commit hooks, conventional commits
• Code quality (axm-audit) — Ruff lint, MyPy strict, test coverage, cyclomatic complexity, security, dead code — composite score on 100 pts

Apache 2.0 · GitHub