dsh-steer-button
Run the following command in DeepSeek Harness:
dsh plugin install lyuwen/dsh-steer-button
Paste the following prompt into your AI chat to install this plugin:
In DeepSeek Harness, you can tell the agent: install plugin https://github.com/lyuwen/dsh-steer-button, or run dsh plugin install lyuwen/dsh-steer-button to install it.
About this plugin
While the agent is mid-task, you want to slip in a note, but the native Harness only offers two coarse options: queue until the whole turn finishes, or cancel and force-send now. There is no middle ground, and queued messages are invisible and uneditable.
dsh-steer-button takes over the composer during a running session and exposes three delivery modes—Queue (the message rides into the next LLM request alongside the current step's observation), Steer (the current step aborts, its partial output is preserved, and the new request continues with your message), Backlog (the message waits for the full turn to finish, then runs as its own turn)—plus a queue strip above the text box that surfaces every pending message with edit, remove, and send-now controls. Pressing the ↑ key pulls the most recently queued text back into the composer for free-form editing.
Built for users who interact with the agent frequently and need fine-grained control over when each message enters the context: neither a blunt interruption nor a passive wait, but a deliberate choice of delivery timing per message.
Use Cases
- Slipping in a quick instruction mid-task without aborting the current step
- Scheduling three messages to land at three different agent steps
- Editing or withdrawing a queued message before it is delivered
Best For
- Developers who interact with the agent at high frequency and care about delivery timing
- Multi-step workflows that require phasing instructions across steps
- Users who want to replace the native busy-state single-key behavior with finer-grained control
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.