Sprig does not try to make coding agents smarter. It tries to give them less to guess. For people and agents; agent-friendly should also mean review-friendly. Experimental Beta; JDK 17+.
Repository automation, data transforms, small JVM applications and compiler utilities. The source milestone's showcases include a repository auditor, a real Maven-library application and a source analyzer. Their READMEs include inputs, commands and explicit boundaries.
The maintainer reports trying a real Sprig workflow with a lower-cost coding model. This is early, anecdotal product dogfooding: there was no controlled experiment, equivalent Java implementation, preregistered task set or productivity metric. It cannot support claims that Sprig beats Java, improves productivity by a measured amount or eliminates model errors. The narrower observation is that compiler queries and structured diagnostics can provide concrete evidence for repairs. We welcome external users to reproduce and report their experience.
Install the SDK or build from source, then:
sprig version
sprig init my-tool
cd my-tool
sprig resolve
sprig runOutput: Hello, Sprig!. Then query sprig capabilities --json, edit src/main.spr, and use sprig check --json to guide repairs.
The published SDK is experimental v0.5.0-beta.1, with compiler version 0.5.0-beta.1 and language version 0.8-dev. Beta brings project testing, wrapper generation, schema-4 dependency locks, explicit Java generic and collection boundaries, and first-party CLI, HTTP, JSON, SQLite and Web packages into the SDK. See release status, the release assets and sprig capabilities --json from the installed SDK for the exact feature set. The end-to-end wiring from a real Fabric/Loom dogfood is in Fabric / JVM framework integration.
Sprig is not self-hosted or production ready. Publishing/registry, LSP, interfaces and generic inference remain future work. See known limitations.
Want to contribute with Codex / Claude / ChatGPT? Pick an agent-friendly issue, read AGENTS.md, run the contributor gate, review the patch and open a PR. Contributing explains the short path; AI assistance is welcome and submitters own review, tests and correctness.
Apache-2.0 · 中文 · Language tour · Tooling and JSON