Recent model updates caused the experimental "sandbox tool assessment"
to time out most of the time leaving the user without any risk
assessment or tool summary. This change explicitly sets the reasoning
effort to medium and bumps the timeout.
This change has no effect if the user hasn't enabled the
`experimental_sandbox_command_assessment` feature flag.
- mark built-in agents that resolve to the current binary as "available" and remove install hints in Agents UI
- thread self-resolved status through overview rendering for consistent labeling in overlay and bottom pane
- add regression test for available status copy
Refs #429
- keep failed overlong coordinator outputs in the conversation, suppress compaction notices, and surface a retry action in the Auto Drive card so follow-up requests retain the error context
- add explicit Action app event, pass compaction notice intent through to the TUI, and raise the coordinator worker stack to 1 MiB with an opt-in backtrace hook for debugging validation overflows
- simplify bottom_spacer_hysteresis_requests_followup_frame to a known viewport pair to avoid FD exhaustion during repeated harness setup
Fixes#432Fixes#430Fixes#433Fixes#429
Replace the retry Thinking event text with a neutral message so the Auto Drive card shows the retry without exposing prompt-length guidance to the user. Build fast remains green.
fixes#432
Add assistant retry note into the conversation so the Auto Drive card shows context while retrying overlong prompt_sent_to_cli errors. Developer guidance string remains in the conversation; raw model output stays in assistant message.
fixes#432
Resolve command check for code/codex/cloud so they are treated as built-in when using the default CLI, even if coder is absent or overridden; prevents skipping agents and avoids running PATH coder when current exe should be used.
fixes#429
James Peter