ax-python-agent-memory-skills

AxAgent Memory And Skills 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

• Load memories or skill guides into an RLM agent run.
• Use static skillsCatalog or memoriesCatalog search without host callbacks.
• Preload constructor or forward-time skills with deterministic id merging.
• Track which memories or skills actually influenced a turn.
• Register non-fatal loaded/used observers in native option maps or target callback wrappers.

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?"})

Lifecycle And State

• Constructor skills seed the loaded-skill prompt without firing load observers.
• Forward-time skills override constructor entries by normalized ID and remain loaded for later calls. IDs and names are trimmed, malformed entries are skipped, valid empty content is preserved, and rendered entries are ID-sorted.
• A forward input named memories seeds the first actor turn. Recalled entries merge by ID for that run, then memory state resets before the next forward.
• onSkillsSearch / onMemoriesSearch take precedence over static catalogs. Without a host callback, skillsCatalog / memoriesCatalog use the built-in deterministic lexical ranker.
• onLoadedMemories / onLoadedSkills observe runtime recall and discovery, not constructor presets. onUsedMemories / onUsedSkills emit one consolidated notification per forward. Forward observers override constructor observers, and observer errors are ignored.
• relevanceRanking produces advisory skill and memory hints using the same tokenization, weighting, tie suppression, limits, snippets, and already-loaded exclusion as TypeScript.
• contextMap is TypeScript's context map: the distiller reads its text (TypeScript's section template when it has none), and after each completed run the context-map Distiller and Cartographer programs update it in place. Since 25.0.0 the text and item ids follow TypeScript's format ([rr-00002], a blank line before each section, a trailing newline); a map saved by a 24.x port loads with its items and is normalized by its next update.
• get_state() and set_state(...) preserve the legacy bare-runtime snapshot shape. Use export_runtime_state() and restore_runtime_state(...) for the complete portable agent snapshot, including loaded skills and constructor-preset reapplication. Do not interchange the two shapes.