[AAASM-4854] 📝 (examples): De-fictionalize production-mode claims in offline examples

[Chisanan232]

What changed

Three of these framework examples advertised a "production / live" mode that the shipped entrypoint does not implement — the src/main.py in each always runs its offline scripted trajectory and never reads a model-provider key, so the documented OPENAI_API_KEY / Gemini-Vertex / model-provider-key claims had zero effect when running the unchanged example. This de-fictionalizes those claims (siblings of AAASM-4837/4838): drop the false key from openai-agents-sdk, and scope the provider-key note in google-adk and smolagents-tool-policy to the "build your own agent" step rather than the unchanged run. A repo-wide sweep (python/, node/, go/ — READMEs + src docstrings + .env.example) turned up one extra instance in the same shape: openai-agents-sdk/README.md's production section prepended OPENAI_API_KEY=sk-... to the unchanged uv run python src/main.py; fixed alongside. Documentation-only — no code behavior changed.

Examples touched:

• python/openai-agents-sdk/src/main.py — docstring: drop the OPENAI_API_KEY production line; use the gateway-URL-only note (matching langchain-basic-agent).
• python/openai-agents-sdk/README.md — retitle "Switching to production mode", move OPENAI_API_KEY off the shipped run and into the "extend main.py" build-your-own step.
• python/google-adk/README.md — move the Gemini / Vertex AI key note from step 2 (.env/run) to step 3 (build a real ADK Agent).
• python/smolagents-tool-policy/README.md — move the model-provider-key note from step 2 to step 3 (build a real smolagents agent).

Kept honest (verified against code, no change): crewai-research-crew & langchain-research-agent (READMEs already state the live path is "not implemented — exits"); all Node examples (explicit "Offline-only demo (no real-provider mode)"); Go examples (scope provider swaps to code changes, not the entrypoint); pydantic-ai & microsoft-agent-framework (already scope the provider key to a build-your-own step); the remaining tool-policy examples make no provider-key-on-entrypoint claim.

Related ticket

Closes AAASM-4854

How to verify

git diff origin/master... shows only Markdown and a Python module docstring — no logic lines. Reading each edited "Switching to production mode" section against its src/main.py: no remaining doc claims that a provider key affects the shipped offline entrypoint. src/main.py still parses (ast.parse); no [tool.ruff]/lint config exists in these examples or at the repo root, so there is no formatter to run for a doc-only change.

Checklist

• PR title follows [AAASM-XXXX] <GitEmoji> (<scope>): <summary>
• No secrets, API keys, or .env files committed
• Example sub-projects include their own README.md with prerequisites and run instructions
• SDK/runtime version dependencies are documented or pinned

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[Chisanan232]

Review — Claude Code

CI: ✅ green — all 16 per-framework example smoke jobs + run + metadata drift + CodeQL + SonarCloud pass.
SonarCloud: 0 open issues.

Scope vs AAASM-4854: ✅ complete. The 3 named fictional production-mode claims (openai-agents-sdk src/main.py docstring, google-adk README, smolagents-tool-policy README) are fixed, plus 1 sweep-found sibling (openai-agents-sdk README). A repo-wide sweep of python/node/go confirmed no other false env/"production"/"live" claims — the honest ones (examples that genuinely read AAASM_GATEWAY_URL, or already scope keys to a build-your-own step) were left unchanged.

Side-effects: doc-only — the diff is Markdown plus one Python module docstring, zero logic lines. The passing per-example smoke jobs confirm every example still runs. No behavior change.

Front-end: none → no Playwright needed.

Verdict: ✅ ready to approve + merge.