Scryer Java Engine

Independent Go wrapper and Kotlin/JVM Java fact producer for Scryer, licensed under Apache-2.0. The engine returns JSON only; terminal, Markdown and HTML reports belong to Core. It imports no Core internal Go packages and uses no Core report assets.

Build and run

Prerequisites: Go 1.24+, Java 21 and Git. The included Gradle Wrapper builds Kotlin.

./scripts/build
./build/bin/scryer-java-engine describe
export SCRYER_ENGINE_JAVA="$PWD/build/bin/scryer-java-engine"
/path/to/scryer project scan /path/to/java-project --static --json
/path/to/scryer project analyze /path/to/java-project --before HEAD~1 --after HEAD

Gradle creates the internal JVM-only intermediate distribution at build/install/java-worker/. Its jars are copied into the complete engine; the Go wrapper starts Java directly and does not invoke the Gradle-generated worker launcher. The installed engine has no dependency on this intermediate directory.

build/bin/scryer-java-engine points into the complete development distribution at build/install/scryer-java-engine/. Keep its bin/ and lib/java/ directories together. Alternatively put the distribution's bin/ on PATH. A standalone go build builds only the wrapper; jars must be supplied via SCRYER_JAVA_CLASSPATH in that case.

describe writes protocolVersion 1, id java, name/version/source and mock false. execute forwards one v1 JSON request from stdin to the JVM and its response back to stdout. Diagnostics use stderr. Both scan and analyze preserve existing Java behavior, including build-tool selection, remote metadata, isolated snapshots, custom test commands, execution/coverage evidence and explicit unknowns.

Runtime and packages

./scripts/package
./scripts/package --with-runtime

Packages contain the Go wrapper, JVM jars, README and Apache license. The second form adds the build host's Java 21 jlink runtime and its legal notices. Runtime selection is SCRYER_ANALYZER_JAVA_HOME, bundled runtime, then JAVA_HOME/PATH. SCRYER_JAVA_HOME independently selects the target project's JVM. Source builds require developer toolchains; a bundled package needs no separately installed Go or Java for the engine itself. Target builds/tests may still need their own tools. GitHub Release publishing and release/source installation remain future work.

Each built/extracted package now includes engine.json (schemaVersion, identity, relative executable, OS and architecture). Install the directory through Core:

scryer engine install java --from /path/to/extracted/package
scryer engine list --json
scryer project scan /path/to/project --static --json
scryer engine uninstall java

Core copies the complete package into ~/.config/scryer by default (including on macOS), with an optional SCRYER_HOME override; installed operation no longer depends on the source checkout or a path override. Existing installs require uninstall before reinstalling. Install/list/uninstall are persistent. The wrapper's package-manifest operation generates metadata during scripts/build; it is separate from describe and scan/analyze fact data. Remote/source installation and GitHub Release publication are still deferred.

Verification and boundaries

The engine CI and Orrery project plan run Go race tests/vet, Kotlin fact/JSON tests, build and both package checks. Core and engine plans run before their workspace integration plan. See AGENTS.md and protocol. Kotlin contains no legacy command/rendering fixtures or MD/HTML report resources.

Current execution is Core → Go wrapper → JVM. The JVM inherits the wrapper's process group so Core can cancel the complete operation on macOS. A future pair experiment will explore a C ABI shared-library wrapper; it is not implemented.