Revisit Python packages and samples after API v2 cutover

[EItanya]

Problem

The Go control plane, CLI, and UI have moved to the API v2 Harness, AgentTemplate, AgentInstance, and upstream A2A model, but python/ packages and samples have not received the same end-to-end audit. Several READMEs and examples still describe Kagent-owned sessions, legacy controller URLs, or persistence behavior that may no longer exist.

Scope

• Inventory python/packages and python/samples by whether each package/sample is supported by the v2 architecture.
• Distinguish framework-internal session terminology from obsolete Kagent session APIs; do not mechanically rename valid Google ADK or framework concepts.
• Update supported packages and samples to use public A2A interaction and current runtime configuration.
• Remove samples and compatibility code whose backing API or service was deleted.
• Verify documented controller endpoints, environment variables, durable-directory behavior, task/context semantics, and deployment manifests.
• Add the smallest runnable validation for each sample we continue publishing, and include it in CI where practical.
• Update Python package READMEs so they do not promise legacy Kagent session persistence or removed REST services.

Done when

Every retained Python sample has an identified supported execution path on API v2, its documentation matches that path, and repository searches no longer find references to removed Kagent-owned session or legacy Agent APIs outside explicitly historical material.

[tryambak2019]

@EItanya
polite request to assign me this issue

[tryambak2019]

Hi @EItanya

kindly refer below the proposed scope for a first, incremental PR so we're aligned before I open one (checked against 2f4e7bd5).

Summary

I checked these README claims against source (not an end-to-end audit) for kagent-openai, kagent-langgraph, and the langgraph/currency sample, and found each one documents an API surface or persistence/task behavior that doesn't match the current code: the kagent-openai README references _session_service.KAgentSession, but that module is absent from kagent-openai at 2f4e7bd5; Quick Start snippets call constructors with kwargs that don't exist; and session/task persistence claims are contradicted by what the code actually does (in-memory task store, no session backend wired up, local-file checkpointing instead of a REST-backed session). Full evidence for each claim - file, line, and the contradicting source — will be included in the PR description rather than here.

Proposed first PR (doc-only, scoped)

• Fix all three READMEs so each code sample matches the real API surface, and each persistence/task-management claim matches current behavior.

What's still pending after this PR

Complete the audit for all packages and samples:

• Inventory of every package (kagent-adk, kagent-core, kagent-crewai, agentsts-adk, agentsts-core, kagent-proto, kagent-skills, and kagent-openai/kagent-langgraph themselves) against the v2 architecture.
• The adk/crewai sample families, the OpenAI basic_agent sample, and the LangGraph hitl-tools/kebab/currency samples.
• Execution tests for retained samples' supported paths and any claimed runtime guarantees.
• Runnable validation; CI where practical.

Does this first increment fit your expectations? I can prepare a draft PR.

[tryambak2019]

Draft PR #2821 is open.

It corrects only the three README files and does not declare any packages or samples fully supported on API v2.

The remaining work includes full package/sample inventory, supported execution-path validation, tests/CI, and removal of obsolete compatibility code where source evidence warrants it.

@EItanya’s scope approval is requested before continuing the remaining multi-step audit.

[tryambak2019]

Follow-up PR #2829 is ready for review. It fixes the published currency console entry-point import and adds CI coverage for ASGI application construction and GET /health with an isolated SQLite checkpoint path.

Together, #2821 and #2829 correct documented legacy behavior and validate one limited currency startup/health slice. They do not close #2689: end-to-end A2A execution, deployment, checkpoint-restart validation, and the remaining Python package/sample inventory are still outstanding.

[tryambak2019]

Current #2689 series status:

• #2821 corrects Python integration documentation that described removed or unsupported behavior.
• #2829 restores the currency package and container entry points, validates main() handing an ASGI app to Uvicorn and GET /health, and confirms the currency image builds. It does not validate live port binding, OpenAI execution, A2A requests, deployment, or checkpoint reuse after restart.

Neither PR closes #2689. Remaining work is LangGraph execution/deployment validation, OpenAI validation, separate ADK and AgentSTS audits/validation, CrewAI audit/validation, and any evidence-backed residual cleanup.

[tryambak2019]

@EItanya @supreme-gg-gg I plan a follow-up PR adding deterministic Python BYO E2E coverage: real A2A invocation through the installed BYO harness plus checkpoint/fork continuity, using a mock model endpoint. Is this the right next increment, and should it target currency or basic_agent?