Is this a reproducible bug?
Current behavior
herdr agent prompt <target> <text> --wait can return success, and --wait can later report the agent as idle/done, while the pane still shows the submitted text sitting unsubmitted in the composer (e.g. /compact never actually starts). Reading the pane shortly after the call still shows the text as the trailing, un-submitted line.
Expected behavior
By the time agent prompt returns (or --wait resolves), the prompt should be fully submitted, or there should be a way to detect/wait for actual submission.
Reproduction
herdr agent prompt <claude-pane> "/compact" --wait --until working then immediately herdr agent read <claude-pane> — the trailing line can still show /compact unsubmitted rather than a started compaction.
Impact
Callers driving agents via the socket/CLI API can't reliably know when a prompt actually started a turn, forcing manual polling workarounds.
Environment
- Herdr version: 0.8.0
- Update channel (stable or preview): stable
- Operating system: macOS
- Terminal: Ghostty
- Shell, if relevant:
- Relevant config, if any:
Is this a reproducible bug?
Current behavior
herdr agent prompt <target> <text> --waitcan return success, and--waitcan later report the agent as idle/done, while the pane still shows the submitted text sitting unsubmitted in the composer (e.g./compactnever actually starts). Reading the pane shortly after the call still shows the text as the trailing, un-submitted line.Expected behavior
By the time
agent promptreturns (or--waitresolves), the prompt should be fully submitted, or there should be a way to detect/wait for actual submission.Reproduction
herdr agent prompt <claude-pane> "/compact" --wait --until workingthen immediatelyherdr agent read <claude-pane>— the trailing line can still show/compactunsubmitted rather than a started compaction.Impact
Callers driving agents via the socket/CLI API can't reliably know when a prompt actually started a turn, forcing manual polling workarounds.
Environment