v1.1.6 (2026-08-19)
Bug Fixes
-
conflicts: Detect Groq behind an OpenAI-compatible base URL (#306,
5aea550) -
conflicts: Do not recover a nested fragment of a truncated reply (#306,
5aea550) -
conflicts: Preserve Groq GPT-OSS conflict output (#306,
5aea550) -
conflicts: Recover conflict JSON that providers wrap in prose (#306,
5aea550) -
conflicts: Request JSON mode and diagnose truncated replies (#306,
5aea550) -
conflicts: Survive real-world conflict-resolution responses (#306,
5aea550) -
examples: Propagate agent-run failures as a nonzero exit (#297,
05f3641) -
guidelines: Guard native tool-call argument formatting against non-dict args (#300,
bfff823) -
guidelines: Parse native tool_calls shape in parse_openai_agents_trajectory (#300,
bfff823) -
guidelines: Stop dropping native tool calls on messages that also carry text (#300,
bfff823) -
guidelines: Warn when removed uncertainty env vars are still set (#300,
bfff823)
Documentation
-
Document custom_llm_provider misrouting risk for mixed-provider resampling (#300,
bfff823) -
Document guideline generation modes and add end-to-end tutorial (#297,
05f3641) -
Fix stale EVOLVE_GUIDELINES_MODE example values (#300,
bfff823) -
Mention both mode alongside regular/consistency in low-code-tracing (#297,
05f3641) -
guidelines: Add missing docstrings on functions touched by this PR (#300,
bfff823)
Features
Detailed Changes: v1.1.5...v1.1.6