What's New in 6.5
Highlights:
- Prompts and answers first. Each turn in the Session view shows the prompt and the agent's answer. Tool calls and working notes now sit behind the Tools chip, off by default; Find still searches them.
- Only typed text is “User.” The label was “You”, since you may be reading someone else's sessions. Notifications, peer messages, commands and scheduled prompts get their own labels; injected instructions and skill files stay hidden unless you turn on "Show harness context in Session view" in Settings.
Bug Fixes:
- Codex sessions with long prompts or instruction wrappers now show a readable, shortened title instead of "Untitled session".
- Codex prompts logged with environment context show again. When Codex wrote the environment block and the prompt as one message, the prompt was hidden from the Text view, export and the session title.
- Copilot sessions on MAI-Code 1.1 Flash show a cost. They had no estimated cost because the model had no price.
- Injected context no longer impersonates your messages. First-prompt navigation also respects explicitly typed instructions instead of treating their wording as a harness preamble.
- Message envelopes read as messages. Known XML wrappers and question replies are decoded; guardian requests show commands and verdicts, and Codex memory citations have a readable footer.
- No more empty agent rows. Empty reasoning placeholders disappear, and consecutive identical messages share one row with a repeat count.
Full release notes: https://github.com/jazzyalex/agent-sessions/releases/tag/v6.5