name: "ax-python-agent-rlm" description: "Use when writing Python code with axllm for RLM executor loops, AxCodeRuntime sessions, runtime envelopes, process runtimes, and optional runtime profiles." version: "25.4.3"

AxAgent RLM Runtime 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

• Wire an AxCodeRuntime or AxCodeSession implementation.
• Use ProcessCodeRuntime or an optional runtime profile for actor-code sessions.
• Explain that generated packages are not TypeScript transpilers; they adapt the Ax runtime contract.

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 agent
helper = agent("question:string -> answer:string")
out = helper.forward(llm, {"question": "How should I proceed?"})

Relevant API Surface

• Agents And RLM: agent, AxAgent, AxAgent.add_child_agent, AxAgent.streaming_forward
• Runtime Profiles: ProcessCodeRuntime, RuntimeCapabilities, RuntimeEnvelope, javascript-quickjs, python-pyodide

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.