pi-working-status

Replace pi's generic Working row with a live phase line. Everything shown is measured from the actual event stream — never estimated:

waiting 4.2s                 request sent → first token (TTFT)
thinking 3.1s · 820 ch       reasoning content streaming
writing 5.4s · 2.1k ch       assistant text streaming
calling write 1.2s           tool-call arguments being generated
bash 12.3s · 1.2k out        tool executing (out = text-block chars emitted)
write 1.4s · 300 out
· quiet 6.1s                 nothing for >3s — stalls become visible

Install

pi install npm:pi-working-status

Then /reload or restart pi.

What the numbers mean

All counts are exact per-delta sums, not estimates:

| Field | Meaning | |---|---| | waiting | time since the provider request went out, until the first delta | | thinking / ch | reasoning characters produced by this LLM call | | writing / ch | assistant text characters produced by this LLM call | | calling | tool-call arguments being streamed (shows the tool name) | | out | characters the running tool has emitted (text blocks only) | | quiet | seconds since the last event of any kind (shown past 3s) |

out counts text only. A tool that returns just images (e.g. read on a screenshot) legitimately stays at 0 — that is a real value, not a hang. Use the trailing quiet timer to tell the difference.

Notes

• No commands, no config: installing the extension turns it on. To disable it, remove the extension (or delete the file before /reload).
• The timer runs at 200 ms; the row only repaints when its text actually changes.
• On agent_settled the row is cleared and the timer stops, so it never leaks into the next turn.

Requirements

• pi on your PATH (Node ≥ 22).

License

MIT