@yusukeshib/pi-agent-controls
v0.1.0
Published
Give pi agents tools to inspect and change thinking levels, compact context, and resume work safely.
Maintainers
Readme
@yusukeshib/pi-agent-controls
A pi extension that gives the agent explicit tools for managing its reasoning level and context.
Tools
get_thinking_level
Returns pi's current effective thinking level.
set_thinking_level
Changes the thinking level used for subsequent model calls. Supported requests are off, minimal, low, medium, high, xhigh, and max. Pi may clamp the request to the active model's capabilities; the tool reports the effective level.
compact_context
Starts context compaction, optionally changes the thinking level afterward, and resumes the session with an explicit continuation prompt. If compaction fails, the continuation is still sent with the error so work can continue without retrying compaction.
The agent is instructed to call this tool by itself because compaction terminates the originating turn, and to persist any exact state that must survive lossy compaction first.
Install
pi install npm:@yusukeshib/pi-agent-controls
# or from git
pi install git:github.com/yusukeshib/pi-agent-controlsReload pi after installation if it is already running:
/reloadLicense
MIT
