Release 0.6.8

[github-actions]

Release readiness review (v0.6.7 -> TARGET e6d2042)

This is a release readiness report done by $final-release-review skill.

Diff

v0.6.7...e6d2042

Release call:

🟢 GREEN LIGHT TO SHIP Changes are largely docs/CI with focused runtime fixes and tests; no blocking risks found.

Scope summary:

• 108 files changed (+3228/-2259); key areas touched: MCP server error handling, DeepSeek reasoning_content conversion, LiteLLM model ordering and dependency bump, CI workflows, doc translations.

Risk assessment (ordered by impact):

1. MCP error handling now wraps HTTP failures into UserError and suppresses some cleanup exceptions.
   • Risk: LOW. Impact: error type/messages may change for callers, but behavior is more precise and still re-raises non-HTTP exceptions.
   • Files: src/agents/mcp/server.py, src/agents/mcp/util.py
2. DeepSeek reasoning_content injection for tool-call messages.
   • Risk: LOW. Impact: only affects DeepSeek/tool-call conversions; guarded by model/provider checks and covered by new tests.
   • Files: src/agents/models/chatcmpl_converter.py, tests/models/test_deepseek_reasoning_content.py
3. LiteLLM optional dependency bump and tool message reordering for Gemini.
   • Risk: LOW. Impact: minor behavior shifts for LiteLLM users; change is targeted to fix API ordering constraints.
   • Files: src/agents/extensions/models/litellm_model.py, pyproject.toml, uv.lock

Notes:

• BASE_TAG from local tags: v0.6.7; TARGET: e6d2042.
• Local verification not run; assume CI has executed the required checks.