skelm, a TypeScript framework for building AI workflows with default-deny security, enters early development

skelm is a TypeScript framework for authoring, running, and operating workflows that mix deterministic code, LLM inference, and agent loops. It runs anywhere Node runs and is currently in early development with unstable APIs until v1.

The framework enforces default-deny security: every agent step declares the tools, MCP servers, network hosts, and filesystem roots it may use, with anything undeclared denied. It includes native support for multiple agent backends (Opencode, ACP, OpenAI, Anthropic, Pi), MCP servers as first-class registry citizens, and per-agent workspaces with persistent audit logs.

Workflows are written as TypeScript modules with three step kinds—code() for deterministic logic, llm() for single inference calls, and agent() for multi-turn loops—rather than a DSL or JSON config. The framework ships with a CLI for scaffolding projects, running workflows once or on schedule (cron, webhook, interval, poll, or queue), and standing up a long-running gateway service.