@howaboua/pi-auto-reasoning-tool

Adds an agent-callable change_reasoning tool. The agent may raise its reasoning effort when the work gets harder, but the user's level at the start of the turn remains the minimum.

Install

pi install npm:@howaboua/pi-auto-reasoning-tool

Behavior

The tool accepts one parameter:

change_reasoning({ level: "low" | "medium" | "high" })

• Requests below the user's turn baseline keep the baseline.
• Users may select xhigh or max; agents cannot select those levels or lower them.
• Pi may clamp a requested level when the current model does not support it. The result reports the applied level.
• After the full agent run settles—including retries, compaction recovery, and queued follow-ups—the extension restores the user's turn baseline.

The model guidance asks agents to change reasoning by work phase, not around individual tool calls.

Local development

bun install
bun run check
pi -e ./src/index.ts