v1.16.20
QwenLM/Qwen3-VLv1.16.20Sep 27, 2026by github-actions[bot]
Upgrading?
AI Summary
Enhances JSON output for session sends to include correlation IDs and fixes several issues regarding Codex session binding and output retrieval.
Key Highlights
- JSON receipts now carry `claude_session_id` and `claude_turn_uuid`
- Codex sessions correctly bind to sibling rollouts and maintain identity
- Improved timeout handling for `session send --wait`
New Features
- Correlation of send operations without separate output queries
- Long or multi-line follow-ups now bound by `--wait`
Full Release Notes
### Added - `agent-deck session send --json` receipts and `session send-status --json` for Claude sessions carry `claude_session_id`, and `--json --wait` also carries `claude_turn_uuid`, the bound user record, so a send can be correlated without a separate output query (#2397, reported by @Abarsit). ### Fixed - Recall indexes profile-scoped Codex configuration directories, including Codex-only profiles without a Claude account slot (#2398, reported by @Abarsit). - A Codex child launched without a message no longer binds a sibling's rollout from the same project: while Codex runs in the pane, only the thread it holds open is bound, so `session output` returns the requested session's conversation (#2394, reported by @Abarsit). - The live Codex identity is persisted at launch, output, stop and archive, so first-turn output is bound and an archived session keeps its `codex_session_id` and `transcript_path` (#2396, #2400, reported by @Abarsit). - `session send --wait` on a Codex session honours the caller's `--timeout` instead of an internal 5 s flush limit that reported a long turn as `completion=timeout` minutes early (#2395, reported by @Abarsit). - Long or multi-line follow-ups to Claude, stored by Claude Code 2.1.277+ as pasted-content rows, are now bound by `--wait` and confirmed by `session send-status`, so the waiter returns after completion instead of holding the full timeout (#2399, #2401, reported by @Abarsit).
Related Repositories
rasbt/LLMs-from-scratch★ 107,072
Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
lyogavin/airllm★ 40,205
AirLLM 2.8T LLM inference with single 4GB GPU
NirDiamant/RAG_Techniques★ 29,960
This repository showcases various advanced techniques for Retrieval-Augmented Generation (RAG) systems. Each technique has a detailed notebook tutorial.
microsoft/OmniParser★ 25,539
A simple screen parsing tool towards pure vision based GUI agent
karpathy/nn-zero-to-hero★ 24,907
Neural Networks: Zero to Hero