AxGen Structured Generation For Java

Use when writing Java code with dev.axllm:ax for AxGen programs, forward calls, indexed multi-sampling, result pickers, streaming, tools, assertions, traces, usage, and output parsing.

Install

Install only this skill for Java:

Shell

npx skills add https://ax-llm.github.io/ax/java/ --skill 'ax-java-gen'

Published skill file: ax-java-gen/SKILL.md.

Source

Source: packages/java/skills/ax-java-gen/SKILL.md
Version: 24.0.17

Skill Instructions

This skill helps an agent write Java code with the generated Ax package dev.axllm:ax. Use the generated package API, examples, and manifests; do not import TypeScript-only APIs unless you are editing the TypeScript package.

When To Use

Build a structured generation program from a signature.
Attach typed tools or MCP-derived tools to a generation call.
Generate multiple validated structured samples and select a winner with a native callback.
Use package examples for no-key scripted clients and provider-api calls.

Package Facts

Language: Java.
Package: dev.axllm:ax.
Package API docs: API.md and axir-api.json.
Capability manifest: axir-capabilities.json.
Runnable examples: examples/.
Real network support: yes.
Scripted no-key transport support: yes.
Runtime profiles: javascript-quickjs, python-pyodide.

Core Pattern

Java

AxGen program = Ax.ax("question:string -> answer:string");
var out = program.forward(llm, java.util.Map.of("question", "What is Ax?"));

Provider Forward Options

AxGen merges constructor and per-call forward options before invoking the provider. Provider-facing keys such as promptCacheKey, sessionId, and contextCache therefore reach the chat request without being copied into program inputs. Per-call values override constructor defaults.

structuredOutputMode / structured_output_mode accepts auto, native, function, or json_object. Auto follows the selected profile/model ordering, with the provider-neutral singleton string/code JSON-object optimization. Explicit modes must be advertised and fail before transport otherwise. JSON-object mode retains exact-shape prompting, strict parsing, and one bounded correction retry without a synthetic __axOutput tool.

Multi-Sampling

Set sampleCount / sample_count to request N provider candidates. Core parses and validates every candidate, preserving each provider result index.
Without a result picker, AxGen returns candidate 0. A result picker receives all { index, sample } structured candidates and returns the winning list index; Core rejects an index outside 0..N-1.
Native callback surface: setSampleCount / setResultPicker.
OpenAI-compatible Chat and Gemini map multi-sampling to n and candidateCount. Anthropic rejects n > 1 explicitly.

Relevant API Surface

AxGen: Ax.ax, AxGen
Tools: Ax.fn, Tool
MCP: AxMCPClient, AxMCPStreamableHTTPTransport, AxMCPStdioTransport

Guardrails

Start from package examples for exact native syntax before inventing a new call shape.
Use provider-api examples only when the user explicitly has provider credentials available.
Use no-key examples for deterministic local checks and provider request mapping.
Treat AxIR as the source of generated package truth: if package docs disagree with source code, update the compiler and regenerate packages.
Do not copy repo-maintainer skills from tools/*/skills/ into user packages.