v0.98.0

[0.98.0] - 2026-02-05

Added

• Exec event parsing for collab_tool_call items (collaboration tool calls).
• Exec event parsing for web_search action payloads.
• Thread option model_personality="none" (mirrors app-server personality support).
• App-server option experimental_api_enabled to opt into experimental methods/fields via initialize.capabilities.experimentalApi=true.
• App-server helpers: thread_name_set, thread_unarchive, thread_compact_start,
• skills_remote_read, skills_remote_write, and skills_config_write.
• thread_list now accepts sort_key and source_kinds.

Updated

• Fixed codex exec argument ordering when resuming a thread with --image attachments
• (resume args now precede image args to avoid greedy flag parsing).
• max_threads validation now only enforces >= 1 (Codex defaults to 6; this is not a hard cap).
• PydanticAI integration updated for pydantic-ai 0.6.x.
• Fixed logfire optional dependency to avoid shadowing Pydantic's Logfire packages in CI,
• which prevented pydantic_ai from importing and caused coverage failures.
• scripts/setup_binary.py now pins the npm download to @openai/codex-sdk@<pyproject version>
• so vendor binaries match the SDK version.
• Bundled Codex CLI vendor binaries updated to 0.98.0 via scripts/setup_binary.py.
• SDK version set to 0.98.0 to match Codex CLI release.

Notes

• Codex 0.98.0 introduces GPT-5.3-Codex (model availability is controlled by your Codex provider).