Release v0.7.9 · yutori-ai/yutori-sdk-python

Quality and hardening release — every finding from a full-codebase review (with multi-agent and Codex cross-reviews) addressed.

Highlights

SDK

py.typed shipped — type checkers now see the SDK's annotations (PEP 561)
Redirects and non-JSON responses raise proper APIErrors instead of silently returning {} or leaking JSONDecodeError; 401/403 errors include the server's detail
API keys are stripped of surrounding whitespace (trailing-newline keys no longer fail deep inside httpx)
extra_body passed to chat.completions.create is no longer mutated
yutori.n1 compatibility restored: submodule attribute access and non-__all__ imports work as they did pre-rename
Navigator payload trimming now converges on multi-image messages; trajectory replay hardened against odd tool-call payloads and escapes screenshot URLs

CLI

Routine API failures print one-line errors with recovery hints instead of multi-screen tracebacks
browse run / research run / scouts create exit 1 when the API rejects the task (scripts no longer see rejected tasks as success)
Output is Rich-markup-safe: queries/URLs containing [...]-style text render literally instead of crashing or disappearing

Installer & packaging

install.sh header version now tracks releases automatically; render dir uses mktemp -d; uninstall.sh accepts "yes"
The sdist ships a complete, collectable test suite; build floor raised to setuptools>=77
New CI: ruff + tests on Python 3.9–3.13, wheel/sdist content checks, twine check, installer freshness — publishing is now gated on tests

docs(llms): be explicit that the agent runs auth and verify itself by @dhruvbatra in #126
docs(llms): CLI-first demos using canonical docs queries by @dhruvbatra in #127
docs: rename install headings and tighten installer section by @dhruvbatra in #128
Fix full-codebase review findings: hermetic tests, py.typed, CLI error handling, installer hardening, CI by @dhruvbatra in #129

Full Changelog: v0.7.8...v0.7.9