[Docs]: Migrate Gemini examples and defaults from 2.5 Flash to 3.6 / 3.5 Flash / Lite

[rosspeili]

Type of fix

Outdated API or CLI behavior

What needs to be fixed?

Skillware examples, skill defaults, and docs still pin Gemini 2.5 Flash (gemini-2.5-flash / gemini-2.5-flash-lite). Google now ships GA Flash models that are better defaults for agent loops and high-throughput evaluators:

Role in Skillware	Old ID	New ID
Default agent / tool-calling examples	gemini-2.5-flash	gemini-3.6-flash
Lite / optional skill evaluators	gemini-2.5-flash-lite	gemini-3.5-flash-lite
Optional mid-tier mention in usage guide	—	gemini-3.5-flash (document as alternate; not required as default)

Explore first (required in PR description): skim current Google / DeepMind Gemini API docs and confirm before merging:

• Model IDs and GA status: Models, 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite
• Function calling / GenerateContent still work with the same google-genai client patterns we use in docs/usage/gemini.md
• Pricing for token_limiter: Gemini API pricing

Expected outcome: changing the model name string is enough — no new Skillware adapter, no new SDK API surface unless Google docs show a breaking change (unlikely for our usage). If anything beyond the model ID is required (config flags, thinking defaults, etc.), document it in docs/usage/gemini.md and apply it consistently.

Replace everywhere gemini-2.5-flash / gemini-2.5-flash-lite appears, including but not limited to:

1. Usage / root docs: docs/usage/gemini.md, README.md, any other usage/agent-loop references
2. Examples: all examples/gemini_*.py, examples/mica_rag_flow.py, examples/build_dataset_demo.py, and any GEMINI_MODEL defaults
3. Skill catalog pages: Gemini Usage Examples under docs/skills/*.md
4. Skill defaults (manifest + skill.py): e.g. compliance/mica_module, compliance/tos_evaluator, wellness/mental_coach (evaluator_model / llm_model defaults and descriptions)
5. Token limiter pricing: skills/monitoring/token_limiter/data/model_pricing.json + docs/skills/token_limiter.md — add/update rows for gemini-3.6-flash and gemini-3.5-flash-lite (and gemini-3.5-flash if listed); update last_updated; remove or keep 2.5 rows only if still useful as legacy aliases
6. Tests / fixtures: e.g. tests/skills/compliance/test_tos_evaluator.py, tests/fixtures/card_ui_schema/monitoring__token_limiter.json — keep assertions in sync with new default strings
7. CHANGELOG.md [Unreleased] (Documentation / Changed)

Non-goals:

• New to_*_tool() adapters
• Rewriting agent-loop logic beyond model ID (and any confirmed API-required knobs)
• Migrating non-Gemini providers
• Mass skill version bumps unless a skill’s public default parameter changes in a user-visible way (then bump patch/minor per CONTRIBUTING)

Acceptance criteria:

• Repo search shows no remaining user-facing gemini-2.5-flash / gemini-2.5-flash-lite defaults (or any leftovers are intentional legacy notes with a clear comment)
• Agent examples and docs/usage/gemini.md use gemini-3.6-flash
• Lite evaluator defaults use gemini-3.5-flash-lite
• model_pricing.json + token_limiter docs match current public list prices (indicative, dated)
• Bundle / maintainer tests that assert model strings still pass
• pytest tests/test_registry_docs.py passes if examples index / catalog pages changed
• PR description cites the Google docs checked and confirms “model ID swap only” (or lists extra knobs if needed)

Suggested verification (offline + optional live):

rg "gemini-2\.5" -g "*.py" -g "*.md" -g "*.yaml" -g "*.json"
pytest skills/monitoring/token_limiter/test_skill.py tests/skills/compliance/test_tos_evaluator.py -q

Optional live smoke (contributor machine, with GOOGLE_API_KEY): run one Gemini example with gemini-3.6-flash and confirm tool calling still works.

Affected Page(s)

docs/usage/gemini.md, README.md, docs/skills/.md (Gemini examples), examples/gemini_.py, skills//manifest.yaml & skill.py (Gemini defaults), skills/monitoring/token_limiter/data/model_pricing.json, tests/ (model string assertions), CHANGELOG.md