gemini-api-dev/examples/README.md

Independent Python request-shape fixture

This fixture executes six preserved Python fenced examples using google-genai 2.29.0 and httpx.MockTransport. Constructor injection, client prefixes, canned responses and mocked sleep are explicit fixture provisions. It serializes native SDK requests and parses synthetic JSON/SSE responses. It never executes a real model, remote agent, repository operation or paid API request.

Eight recorded runs cover quickstart, a previous_interaction_id link, completed and failed polling, bounded cancelled polling, remote environment request shape, custom agent request shape and text/usage SSE parsing. The cancelled Python polling example does not stop on cancelled; the fixture stops it after two mocked responses. The original example is preserved. This is an observed counterexample, not an instruction to run an unbounded polling loop against a paid service.

Reproduction from the extracted gemini-api-dev folder:

python -m venv .fixture-venv
# Activate that dedicated virtual environment using your platform's method.
python -m pip install -r examples/installed-distributions.txt
python examples/run_fixture.py

New reproduction output is written to examples/output, preserving the shipped fixture-evidence.json and distribution record.

Installing dependencies contacts a public package registry. No real API key is needed by the fixture. Use a dedicated environment; do not import credentials or cookies. The distribution list records the tested environment, but is not a cross-platform lock with hashes. Review wheels and platform requirements first.

The runner uses sockets/DNS denial and an injected MockTransport. The eight-run record is evidence of limited SDK behavior, not service-side authentication, model availability, billing, context memory or full AI-client skill execution. TypeScript, multimodal, TTS, real-time Live API and deployment remain untested.

The original skill recommends current model names, while parts of the retained migration guide contain different legacy-model advice. Recheck official model and deprecation documentation before applying those recommendations. The model IDs in these synthetic requests are example strings, not availability proofs.

BB Skills additions are covered by examples/LICENSE.BB-SKILLS.txt. Upstream instructions and documents remain Apache-2.0 with their original notice.