v0.7.4

v0.7.4

Highlights

• Canonical provider surface is now gemini for the direct Gemini API, gemini-cli for the Gemini CLI, and vertex-ai for Vertex Model Garden access.
• CLI provider handling is now provider-specific for Claude Code, Codex, and Gemini CLI instead of assuming one shared JSON/output contract.
• Large bounded runs now expose better preflight and per-phase timing data, and same-provider calls are throttled across concurrent council processes.
• Vertex Gemini now defaults to the global location rather than the stale us-central1 default.
• Packaged docs, repo skills, and Claude plugin metadata are refreshed for the current 0.7.4 surface.

Validation

• ./.venv/bin/python -m pytest -q → 298 passed, 2 warnings
• ./.venv/bin/python -m compileall src tests → passed
• ./.venv/bin/python -m build → built the_llm_council-0.7.4.tar.gz and the_llm_council-0.7.4-py3-none-any.whl
• Sequential live smoke passed for active API-backed providers:
  • openai
  • anthropic
  • gemini
  • vertex-ai

Known issue

• gemini-cli deep doctor passes on the local Vertex-authenticated setup, but review-mode smoke remains unstable under council orchestration. It can overrun timeout budgets or return empty non-JSON output. This release does not claim that gemini-cli review-mode automation is fully stable yet.