Ax Playbook 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

• Grow an evolving context playbook for a program or agent stage with playbook().
• Attach a seed playbook to an agent and learn bounded avoidance rules from run-end failure signals.
• Use the agent-bound playbook evolve method to mine grounded weaknesses with verification and exact rollback.
• Refine a playbook online from live feedback or offline from labeled examples.
• Render or persist a playbook and inject it into a program context.

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

python
from axllm import ax, playbook

program = ax("question:string -> answer:string")
pb = playbook(program, {"studentAI": llm})
pb.evolve(examples, metric_fn)

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.