[codex] Expose standalone typed tools

[anth-volk]

Fixes #93

Summary

• Register the standalone typed tools with the chat runtime: calculate_household, run_economy_simulation, and analyse_microdata.
• Update prompt/tool guidance so typed tools are preferred when their shape fits, with run_python kept as the fallback for structural reforms, introspection, and unsupported cases.
• Block FRS row-level access in analyse_microdata while preserving aggregate FRS outputs through run_economy_simulation.
• Add frontend labels, runtime docs, and focused tests for tool exposure, FRS rejection, mocked non-FRS microdata behaviour, and prompt contracts.
• Default chat temperature to 0 and keep model-facing tool-result ordering deterministic while streaming frontend results as tools finish.

Validation

• PYTHONPATH=backend python -m pytest backend/tests/test_agent_tools.py backend/tests/test_prompts.py backend/tests/test_structural_tools.py
• python -m py_compile backend/tests/test_agent_tools.py backend/agent_tools.py backend/routes/chatbot.py
• git diff --check

Notes

This is stacked on #92 to keep the standalone-tool review scoped to the 8-file diff on this branch. If preferred, it can be retargeted to main, but that would include the prompt/testing base from #92 in this PR as well.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
policyengine-uk-chat	Ready	Preview, Comment	May 29, 2026 11:41pm

[github-actions]

Beta preview is ready.

• Frontend: open preview
• Backend: open backend