Beta (v0.7.2-beta.14.1.25324db)

Automated beta build from main (25324db1f35f6cdcf24ded7359eaf3e33d26d176).

v0.8.0

• Added install.ps1 for a one-line Windows install (irm ... | iex), matching install.sh; trimmed the README's Getting Started section to two lines.
• Fixed the automatic uv installer always running the POSIX sh -c "curl ... | sh" form, which threw ENOENT on Windows; it now runs the real Windows installer via PowerShell.
• Fixed pi.exec() (the extension SDK's exec helper) always spawning with shell: false, breaking .cmd/.bat-shimmed binaries on Windows.
• Fixed $VISUAL/$EDITOR being split on a bare space, which shredded a quoted Windows editor path.
• Added rlm.set_model(model=..., model_class="same"|"smaller") so the running agent can change its own active model mid-session, reusing the same classification-based ranking rlm.run(modelClass=...) already uses for subagents.
• Fixed advisor.consult() returning an unusable status by renaming the reply's status field (which collided with the kernel host bridge's own reserved status key) to outcome.
• Changed the advisor's default reviewer model from the session's own model to the cheapest authenticated model in classification class "S", so a second opinion actually comes from a different, stronger-tier model by default.
• Added ask-when-ambiguous and proactive-compaction guidance to the core system prompt so the ask-user and compact skills are actually used as intended, not just discoverable.
• Changed network-error auto-retries to say "Reconnecting" instead of the generic "Retrying" in the interactive UI.
• Added a crash handler for the foreground/interactive client (the daemon already had one), logging uncaught exceptions/rejections to client-errors.log instead of losing them.
• Added rlm.log_error(message, **context) so the agent can add its own findings to the same structured error log.
• Enhanced zero doctor to surface recent errors from the structured and client error logs, not just background service status.
• Changed the subagent status line below the chat input to expand into a per-subagent list inline on Enter, instead of opening a separate Session View screen.
• Removed build-from-source and pinned/beta install instructions from the README; the GitHub Release install path is now the only documented one.
• Changed the agents view splash hint from "type to start" to "type to search sessions".
• Added app.edits.expand (ctrl+j) to toggle edit diffs; diffs are now shown only by this toggle, and ctrl+o no longer affects them.
• Changed edit rendering so the ╰─ <path> +N -M summary line is always visible and ctrl+j toggles the diff inline beneath it, indented to the summary text.
• Fixed fullscreen wheel scrolling in Ghostty while retaining application link clicks; set terminal.fullscreenMouse to false to use native Cmd-click instead.
• Changed the agents view to sort idle and inactive sessions by last message time, newest first, while keeping running agents in stable creation order.
• Fixed openai-codex models being invisible to rlm subagents and find_models because model discovery reported Prime Agent's own version as the Codex client version (#1375 by @bilelrais).
• Added a working hint that recommends sharing traces with Prime Intellect to help train open-source LLMs.
• Restored bare prime-agent --resume opening the agents view and the /resume [id|path] slash command; bare commands open the agents view and an argument resumes that session in place.
• Fixed URLs not opening on click in fullscreen mode on terminals such as Ghostty; clicking a link in the transcript, dock, or overlays now opens it in the browser.
• Fixed ctrl+p ("Toggle agent message expansion") only toggling received agent messages; it now expands and collapses sent agent messages together with received ones.

v0.7.3

Release v0.7.3

v0.7.2

• Fixed Down Arrow focusing the Agents View entry before moving a nonempty prompt cursor to the end (ENG-5147).
• Added app.messages.expand (ctrl+p) to collapse or expand agent-to-agent messages separately from ctrl+o tool output.
• Added a ctrl+t expand hint to collapsed thinking blocks, matching the tool output hint.
• Changed expand/collapse hints to a consistent bracketed (Ctrl+O to expand) style across tool, message, summary, and error rows.
• Added a configurable copy action to login dialogs so raw sign-in URLs can be copied without selecting wrapped text (#643).
• Added privacy-safe pseudonymous product analytics for onboarding, command use, execution modes, run outcomes, TTFT, latency, usage, tools, retries, and compactions, with disclosure and opt-out controls (ENG-4682).
• Changed sent agent messages in the IPython cell UI to show only the message text with a ╰─ gutter when expanded, matching received messages, and hid the raw agent_message.send receipt dictionary.
• Fixed Homebrew installs attempting to self-update their versioned Cellar keg instead of directing users to brew upgrade prime-agent (#844)
• Fixed the agents view collapsing expanded subagent lists when returning from an opened agent (ENG-5105).
• Kept the subagent summary row visible and selectable while its list is expanded in the agents view, so pressing enter on it collapses the list again (ENG-5105).
• Added in-place editing of queued steering and follow-up messages: Alt+Up/Alt+Down browse the queue from the draft, Enter applies the edit as steering, Alt+Enter as a follow-up, and submitting an empty editor deletes the item; interrupts now preserve the queue (#838).
• Fixed workers with no live connection reporting as ready; stopping workers now report a stopping state, are hidden from live sessions, and no longer receive daemon-wide commands (#850).
• Fixed timed-out worker stops stranding dead-but-registered workers ("Session worker is not connected"); stops now finalize in the background once the process exits, and zombie processes are no longer counted as alive (#851).
• Fixed sessions becoming permanently unopenable after a stale worker registration was left behind; open/resume now self-heals by finishing the old cleanup and starting a fresh worker (#852).