foundr.work
Write the spec. The framework ships the worker.
Foundr Workers are stateful agents you describe in a YAML file. The foundr-work npm package and a bundle of skills for Claude Code or Codex compile the spec into a worker that runs long-running tasks against your product requirements.
What it is
Foundr.work is an opinionated framework. You declare a worker — its objective, its tools, its state, its escalation rules — in a YAML file. The foundr-work npm package and a bundle of skills for Codex or Claude Code compile that spec into a stateful agent that picks up long-running tasks and ships them.
How it works
4 steps, in order.
- 01
Declare
Write a YAML spec: the worker's job, its tools, its state shape, and its escalation rules.
- 02
Compile
Run the foundr-work CLI; it scaffolds the worker using the skill bundle in your Codex or Claude Code config.
- 03
Run
The worker runs locally or on hosted runtime, picks the next task, ships, persists state.
- 04
Iterate
Tweak the YAML, recompile. Workers are versioned by spec hash, not by code drift.
What makes it different
Declarative, not imperative
The spec is the source of truth — you never write the agent loop.
Stateful by default
Workers persist across runs, restarts, and model swaps.
BYO CLI
Skills work with Codex or Claude Code; the worker doesn't care which.
Why it exists
Most agent frameworks make you write Python and pray. Foundr Workers are a contract: write the YAML, get a stateful agent. The framework is the spec, not your code.