Ax Python Gepa

Ax GEPA For Python

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

When To Use

• Run the generated GEPA optimizer or inspect a GEPA artifact.
• Use BootstrapFewShot before GEPA when demonstrations should seed optimization.
• Track metric budgets, reflection calls, candidate state, and Pareto fronts.

Package Facts

• Language: Python.
• Package: axllm.
• 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

from axllm import AxGEPA

engine = AxGEPA(reflection_client)
result = engine.optimize(request, evaluator)

Relevant API Surface

• Optimizers: optimize, playbook, AxPlaybook, AxBootstrapFewShot, AxGEPA, OptimizerEngine, OptimizerEvaluator

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.