You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.Y
8000
ou switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
boot-agents.sh reports "launched + armed" while the pane is waiting on a Claude Code prompt #24
A new agent's pane stops at Claude Code's first-run screens: folder trust (unnumbered, default selection is "No, exit"), then the resume picker (see the linked issue on that), then the fullscreen-renderer offer. The script still reports success, so a boot-on-startup produces no working agent and the user is not told. Prompts change between Claude Code versions, so they cannot be enumerated once.
Evidence
Seen in test: multiple runs stalled at the trust screen while the boot script reported "1 launched + armed".
Fix
After launching, sample the pane (reuse the pane-idle / stuck-pane classifiers) and report per agent: ready, or "waiting on a Claude Code prompt: open its terminal in Deck". Recognise the trust screen specifically and say "choose Yes, I trust this folder; do not press Enter, No is preselected". Document this in INSTALL.md tier 3, including that agents run in Claude Code's default permission mode (currently auto) unless configured. Do not write into Claude Code's internal trust config without a decision; it is version-fragile.
Acceptance criteria
With a fresh agent, boot-agents.sh exits with a summary naming each agent stuck at a prompt, and a non-zero or distinct exit status for "launched but waiting".
INSTALL.md tier 3 has a "first boot" paragraph.
Test covers a pane fixture showing the trust screen.
Activity
docs(base): record PR 54 review feedback in spec (#24)
fix(base): poll readiness in rounds and report refused launches (#24)
fix(base): honest launch status (#24) (#46)