Repository navigation
Replies: 2 comments 1 reply
|
👍 I use herdr to orchestrate subagent panes from a pi extension. The extension spawns wrapper scripts that manage the agent's launch environment and lifecycle. Until 0.7.5 this worked via agent start -- which was completely reliable. The new model requires coordinating splitting a shell pane and waiting for shell init before typing a command into it. That introduces race conditions that direct argv spawn avoided. There's no way to reliably know when an arbitrary shell environment (direnv, nix, slow rc) is ready to accept input. @etofsa I've been using this branch as a workaround: https://github.com/elucid/herdr/tree/pane-split-argv. I was going to open the same issue and found yours. |
|
Another use case, from the socket API side. I'm building a pi extension that runs subagents in herdr panes. A tab can run an exact argv through
A split that launches an argv directly, the way |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
idea / problem
I use Herdr socket automation to create visible panes for long-running tools.
pane.splitalready lets a client choose the workspace, target pane, direction, ratio, cwd, environment, and focus behavior, but the new pane always starts the normal shell.Launching the intended process then requires a second
pane.runor input step with a command string. That loses argument boundaries and introduces shell quoting and expansion where the caller already has a structured argv vector.requested change
Allow
pane.splitto accept an optional non-emptyargvarray:{ "id": "split-worker", "method": "pane.split", "params": { "target_pane_id": "w1:p1", "direction": "right", "cwd": "/path/to/project", "env": {"ROLE": "worker"}, "argv": ["python3", "worker.py", "--label", "review queue"] } }When
argvis omitted, behavior stays exactly as it is today. When present, Herdr launches that vector directly without a shell, preserving every argument literally and returning the normal pane identity. Empty or invalid vectors should fail clearly.This can reuse Herdr existing argv-based terminal launch path. It does not require changes to
agent.start, persistence, layout identity, UI behavior, or the socket authority model. I have validated the shape in a small isolated prototype and would be happy to contribute it if the direction fits.why you want this
It gives automation a deterministic way to create a visible, controllable pane around an exact process. Agent orchestrators, test runners, and other tools can avoid constructing shell commands while retaining Herdr layout, observation, and lifecycle controls.
All reactions