feat(panel): --allow-fresh flag + fix cube feedback stdin double-read

[jacsamell]

Two fixes batched into one PR.

1. --allow-fresh CLI flag on cube peer-review and cube pr-review (alias prv). Soft-fall-through to fresh for judges with missing/stale sessions instead of hard-crashing the whole panel. Default stays hard-crash so cube state-management bugs surface loudly; operators opt in when they already know the sessions are dead. Plumbs through launch_judge_panel → peer_review_command → _run_pr_review → _run_branch_review.

2. cube feedback /dev/stdin <<EOF no-op fix. Typer consumed stdin during arg parsing; feedback_path.read_text() then saw EOF and sent codex an empty prompt. Special-case /dev/stdin and - to drain sys.stdin once into a temp file. Errors loud on empty input.

Test plan

• cube prv <pr> --allow-fresh against a task with deleted/stale session files — proceeds with that judge fresh, others resume
• cube prv <pr> without the flag — still hard-crashes on missing session (default behavior preserved)
• cube feedback writer-a my-task /dev/stdin <<EOF ... EOF — codex actually receives the heredoc content
• All 237 tests pass (verified locally)

Overview

This PR adds two independent features to improve panel reliability and feedback UX:

1. --allow-fresh Flag

Adds optional --allow-fresh to cube peer-review and cube pr-review (alias prv) commands. When enabled, judges with missing or stale session files soft-fall through to launching a fresh instance instead of hard-crashing the entire panel. Default remains strict (hard-crash) to surface state-management bugs; operators opt in when expecting dead sessions. The flag threads through: CLI → peer_review_command / pr_review_cmd → _run_pr_review / _run_branch_review → launch_judge_panel.

2. Feedback stdin Double-Read Fix

Fixes silent no-op when running cube feedback writer-a task /dev/stdin <<EOF ... EOF. Typer's argument parsing consumes stdin during setup, leaving subsequent reads seeing EOF. Now special-cases /dev/stdin and - to drain remaining sys.stdin into a temporary file under .prompts/, with loud error on empty input to surface the issue immediately.

Files Modified

• judge_panel.py: launch_judge_panel() gains allow_fresh parameter with conditional soft-fall-through logic
• cli.py: Adds --allow-fresh option to peer-review and pr-review command definitions
• peer_review.py: Threads allow_fresh through peer_review_command(), _run_pr_review(), _run_branch_review()
• feedback.py: Detects stdin paths and buffers content to temp file before processing

Test Coverage

Existing test suite (237 tests) passes. Test test_run_judge_retries_fresh_when_saved_session_missing validates judge retry logic with missing sessions.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 463c1c16-d717-43bd-b97d-a599bd557262

📥 Commits

Reviewing files that changed from the base of the PR and between a3ed1c0 and 5495f6c.

📒 Files selected for processing (4)

• python/cube/automation/judge_panel.py
• python/cube/cli.py
• python/cube/commands/feedback.py
• python/cube/commands/peer_review.py

---

Walkthrough

This pull request introduces an allow_fresh parameter to gracefully recover from stale judge sessions during panel resume, adds CLI options and end-to-end wiring to expose this feature, and separately adds stdin support to feedback submission.

Changes

Allow-fresh graceful judge panel resume

Layer / File(s)	Summary
Judge panel allow-fresh fallback logic python/cube/automation/judge_panel.py	launch_judge_panel now accepts allow_fresh: bool = False and conditionally starts a judge fresh (by clearing session_id) when session loading fails, instead of raising immediately.
CLI option declaration for peer-review and pr-review commands python/cube/cli.py	New --allow-fresh boolean option added to both peer-review and pr-review commands with descriptive help text. The pr-review docstring is updated with a usage example.
CLI argument propagation to peer_review_command python/cube/cli.py	All peer-review and pr-review CLI code paths (GitHub PR mode, local/branch mode, and default PR mode) now pass allow_fresh through to underlying peer_review_command calls.
Command layer threading through peer_review_command and sub-functions python/cube/commands/peer_review.py	peer_review_command exposes allow_fresh and forwards it to _run_pr_review, _run_branch_review, and all launch_judge_panel calls across single-judge, fresh, and resume execution paths.

Stdin feedback submission support

Layer / File(s)	Summary
Stdin input and temporary file handling in feedback_command python/cube/commands/feedback.py	When feedback_file is /dev/stdin or -, the function reads remaining stdin content, validates it is non-empty, writes it to a temporary markdown file under .prompts/ using task_id, and uses that path for submission. Non-stdin inputs follow existing resolve_path() logic.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Possibly related PRs

• aetheronhq/agent-cube#177: Modifies session resume fallback in judge_panel.py to tolerate missing judge sessions instead of crashing, complementary to the new allow_fresh graceful handling.
• aetheronhq/agent-cube#175: Updates the PR peer-review workflow in peer_review.py to verify worktree HEAD matches PR head, affecting the same judge panel launch entry points.
• aetheronhq/agent-cube#153: Extends the shared peer_review_command interface in the same CLI and command modules, adding parameter-threading patterns similar to allow_fresh.

Poem

🐰 A rabbit's ode to fresh starts

When sessions grow stale and panels won't wake,
A gentle fallback is all that it takes—
With allow_fresh set, judges start anew,
And feedback flows freely through stdin too! 🎯

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}