@yusukeshib/pi-working-status

A pi extension that replaces the plain Working... loader with a live, detailed status line tracking the agent through every phase of a turn — request, streaming, and tool execution.

What it looks like

The detail in parentheses changes as the turn progresses:

Working... (0:02 · sending request to api.anthropic.com…)
Working... (0:04 · receiving reasoning stream… 240 tok · 1.2 KB)
Working... (0:07 · receiving text stream… 1.1k tok · 180 tok/s · 4.8 KB)
Working... (0:08 · ⚙ bash: npm test -- --watch=false)
Working... (0:09 · ⚙ running bash, read)

A fixed elapsed clock follows Working, and a detail segment shows the most specific current activity:

| Detail | Meaning | | ------------------------------------- | -------------------------------------------------- | | ⚙ <tool>: <args> | single tool executing (command / path / pattern) | | ⚙ running <tools> | multiple tools executing (deduped names, max 3) | | sending request to <host>… | request being sent (first ~0.4s) | | waiting for response… <n>s | request sent, awaiting any response | | receiving reasoning stream… | provider streaming a reasoning block | | receiving text stream… | provider streaming assistant text | | receiving tool call… | provider streaming a tool call | | connected to <host>, waiting… <n>s | headers received, awaiting first token | | HTTP <status> | last response returned a non-2xx status |

Streaming lines append token counts, tok/s, and decoded byte counts when the provider reports them mid-stream (Anthropic, Bedrock, Mistral, Google). Model id and context-window usage are intentionally omitted — pi's footer already shows them.

Install

pi install npm:@yusukeshib/pi-working-status

Via git (no npm publish required):

pi install git:github.com/yusukeshib/pi-working-status

Try it without installing:

pi -e npm:@yusukeshib/pi-working-status

Command

/working-status [on|off]

• /working-status off — restore pi's default Working... message
• /working-status on — re-enable the detailed status line
• /working-status — show the current on/off state

How it works

The extension subscribes to pi's turn lifecycle events (before_provider_request, after_provider_response, message_update, tool_execution_start/end) and a 1s ticker, then composes the status line through ctx.ui.setWorkingMessage(). Token counts come from the provider's streamed usage; tok/s and time-to-first-token are derived locally. Byte counts are decoded payload size, not raw HTTP bytes.

License

MIT