axonpush
Getting started

Supported frameworks

The 23 sub-skills in the axonpush bundle, agent frameworks, OpenTelemetry exporters and log forwarders, what each one wires up, and how detection picks between them.

axonpush/skills ships one orchestrator and 23 sub-skills. Each is a host-agnostic Markdown bundle that installs the right SDK extra and wires the integration into your entry point, and each is invocable on its own, /langchain, /ts-vercel-ai, /pino, as well as through /axonpush-integrate.

The skill slugs are bare names. There is no axonpush- prefix on the sub-skills; only the orchestrator carries it.

Agent frameworks, Python

SkillSDK extraReference
/anthropicaxonpush[anthropic]Anthropic
/crewaiaxonpush[crewai]CrewAI
/deepagentsaxonpush[deepagents]Deep Agents
/langchainaxonpush[langchain]LangChain
/openai-agentsaxonpush[openai-agents]OpenAI Agents
/otel-pythonaxonpush[otel]OpenTelemetry
/customaxonpushOverview

/langchain covers LangGraph as well. /custom is the escape hatch: direct events.publish calls for a framework nothing else handles.

Agent frameworks, TypeScript

Every one of these installs @axonpush/sdk; the integration lives at @axonpush/sdk/integrations/<name> and the framework is a peer dependency.

SkillReference
/ts-anthropicAnthropic
/ts-google-adkGoogle ADK
/ts-langchainLangChain
/ts-langgraphLangGraph
/ts-llamaindexLlamaIndex
/ts-mastraMastra
/ts-openai-agentsOpenAI Agents
/ts-vercel-aiVercel AI SDK
/otel-tsOpenTelemetry
/ts-customOverview

Log forwarders

These do not instrument an agent. They point the logging library you already use at axonpush, so your application logs land on the same timeline as your agent events, as app.log events joined by the same trace ID. Wiring one of these alongside a framework skill is the normal configuration, not an alternative to it.

SkillLibraryReference
/loggingPython stdlib logging, including a Django dict-configlogging
/loguruLoguruLoguru
/structlogstructlogstructlog
/pinoPinoPino
/winstonWinstonWinston
/consoleNode console captureconsole

None of them replaces your logging library, each attaches a handler, sink, processor, stream or transport to the one that is already there.

Detection and selection

/axonpush-integrate scans:

  1. pyproject.toml, requirements.txt and package.json for dependency names.
  2. Source files for import statements.

You can pick more than one. The orchestrator presents everything it found and lets you select several, an agent framework plus a log forwarder is the common case, and multiple frameworks are supported. It does not force a single choice.

Two detection details worth knowing:

  • Google ADK is detected from @google/genai or @google/generative-ai, not from a package called google-adk.
  • The detector also reports bunyan and Python’s bare print, neither of which has a matching skill. The orchestrator drops anything it cannot map, so those simply do not appear as options. For print, the Python SDK has setup_print_capture, wire it by hand.

Invoking a sub-skill directly overrides detection entirely.

Not covered: .NET

There is no C#, Semantic Kernel or AxonPush.Otel skill, and the detector reports only python, typescript, both or unknown. Install and wire the .NET packages yourself, see the .NET SDK reference.

Staying in sync

Each framework sub-skill fetches the matching section of the live SDK README at run time and uses the static code in its SKILL.md only as an offline fallback. What the agent writes tracks the SDK, not the version of the bundle you installed.

Source

github.com/axonpush/skills, the skills are plain Markdown, so reading one before you run it takes about a minute.