v2.0.0-alpha.3

What's Changed

• Feature/structured streamable output by @timazed in #7
• Add queryable history, hidden compaction, and runtime observation by @timazed in #8

Full Changelog: v2.0.0-alpha.2...v2.0.0-alpha.3

v2.0.0-alpha.2

What's Changed

• Feature/structure output new demo app UI by @timazed in #4
• Add layered memory capture APIs and demos by @timazed in #5
• Split runtime and backend files by concern by @timazed in #6

Full Changelog: v1.1.0...v2.0.0-alpha.2

v1.1.0

Added

• Runtime skill support with thread-pinned skills, per-turn skill overrides, and execution-policy enforcement for allowed tools, required tools, tool sequence, and max tool calls.
• Dynamic persona and skill loading from local files and remote URLs through AgentDefinitionSource.
• Resolved-instructions preview support so host apps can inspect the final compiled instructions for a turn.
• Transient request retry/backoff policy support in the Codex responses backend.
• Configurable reasoning effort (low, medium, high, xhigh) for gpt-5.4 style thinking control.
• Demo app UI for switching thinking level on future requests.

Changed

• Added CLI-style unauthorized-session recovery so runtime operations can refresh and retry once after auth expiry or invalidation.
• Improved the demo app’s skill and Health Coach flows to better show persona, skill, and tool orchestration together.
• Expanded README coverage for retries, skills, dynamic definition sources, and reasoning effort configuration.