Feat/google adk polish

[amanishimwe]

Pull Request

Description

Polishes the Google ADK integration for TealTiger: exports TealTigerCallback, adds Gemini pricing entries, wires model-based tool cost estimates (with fallback), and adds unit tests plus a runnable example.

Fixes #324

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Performance improvement
• Code refactoring
• [x ] Test improvements

Changes Made

• Export TealTigerCallback from tealtiger. integrations
• Add Gemini model pricing (gemini-2.5-flash, gemini-2.5-pro, gemini-3.5-flash, gemini-3.6-flash)
• Estimate ADK tool cost from get_model_pricing (fixed 500/500 token estimate); fall back to cost_per_tool_call when pricing is missing
• Add examples/google_adk_governance.py (mock before_tool / after_tool demo, no API key required)
• Add/extend tests in tests/test_google_adk_integration.py and Gemini pricing coverage in tests/cost/test_pricing.py

Testing

Describe the tests you ran to verify your changes:

• Unit tests pass (pytest)
• Type checking passes (mypy src)
• Linting passes (ruff check src tests)
• Manual testing performed
• Added new tests for new functionality

Checklist

• My code follows the project's code style
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published

Breaking Changes

None

Screenshots (if applicable)

N/A

Additional Notes

• Tool cost is an estimate (500 input + 500 output tokens) when the model exists in the pricing table; it is not live token usage from ADK.
• A follow-up contribution to Google ADK contributing is planned separately (CLA required).
• before_tool retains # noqa: C901 for pre-existing complexity.

[amanishimwe]

@nagasatish007 , when you get time, please look at this one as well. A follow-up contribution to Google ADK contributing is planned separately (CLA required)

[nagasatish007]

Thanks @amanishimwe — good work on this. The scope is clear and the PR description is thorough.

Looks good:

• Export wiring for TealTigerCallback from integrations ✅
• Gemini pricing entries (2.5-flash, 2.5-pro, 3.5-flash, 3.6-flash) ✅
• Runnable example with no API key dependency ✅
• Test coverage for both integration and pricing ✅

Questions/suggestions before merge:

1. Fixed 500/500 token estimate for tool cost — is there a way to pull actual token counts from the ADK callback context (before_tool/after_tool)? Even an approximate count from len(str(args)) would be more accurate than a static 500. If ADK doesn't expose this, the fixed estimate is fine for now but let's add a # TODO: use actual token counts when ADK exposes them comment.

2. CI status — showing 5/21 checks passing. Are the failures pre-existing (like the TypeScript repo's CostStorage issue) or related to this PR? If pre-existing, note that in a comment so we can track separately.

3. # noqa: C901 in before_tool — understood it's pre-existing complexity. If you have bandwidth in a follow-up, would be good to refactor that method (extract helper functions for policy evaluation vs. cost tracking). Not blocking this PR though.

4. Gemini model naming — are gemini-3.5-flash and gemini-3.6-flash confirmed production model names from Google? Want to make sure we're not adding pricing for beta/unreleased model IDs that might change.

Overall this is ready to merge once CI is clarified. Nice contribution 👍

[nagasatish007]

@amanishimwe Also please join our Discore server : https://discord.gg/ZnCEd5Mk2

[amanishimwe]

@nagasatish007 Thanks for the review — I have addressed your comments as follows:
Token estimate
Updated cost estimation to use an approximate count from len(str(payload)) // 4 when args/result are available, with 500 kept only as a fallback when they’re not. Added # TODO: use actual token counts when ADK exposes them.
# noqa: C901 in before_tool
Refactored policy evaluation into helpers (_check_freeze, _check_tool_allowlist, _check_pii, _check_secrets, _check_cost_limit) and removed the noqa.
Gemini model naming
Confirmed gemini-3.5-flash and gemini-3.6-flash are production/stable model IDs in Google’s Gemini API docs
CI
These are pre-existing and not related to my changes

[amanishimwe]

@nagasatish007 if you get time, look at this PR.

[nagasatish007]

@amanishimwe — reviewed the latest changes. All four items from my earlier review are addressed:

1. ✅ Token estimate improved with len(str(payload)) // 4 + TODO comment for when ADK exposes actual counts
2. ✅ before_tool refactored into clean helper functions — much better
3. ✅ Gemini model names confirmed
4. ✅ CI failures noted as pre-existing

One note: the import sorting commit (c9deb56) touches 74 files that aren't related to the ADK feature. Not blocking, but in the future it's cleaner to keep style-only changes in a separate PR — makes review easier and git blame cleaner.

Merging this. Nice work 👍