@arvoretech/pi-recap

Drafts a one-line recap of where you left off in a Pi session — inspired by Claude Code's session recap.

What it does

• On demand: run /recap any time to get a 1–2 line summary of what was being worked on, what's done, and the next pending step. Useful before a commit, before opening a PR, or after a long session.
• Automatically on return: when you come back to a session that has been idle for a while, Pi drafts the recap and shows it before your next turn — so you don't have to scroll back to rebuild context.

The recap is generated by a background model call (via the host's completeSimple, same mechanism as smart-context) using the current session model. It's best-effort: if it can't run, it silently no-ops and never blocks your turn.

Auto-recap trigger

Pi has no native terminal focus/blur event, so instead of Claude Code's "unfocused ≥ 3 min" signal this extension uses idle wall-clock time since the last completed turn:

1. At least 3 minutes elapsed since the last agent_end.
2. Session has at least 3 user turns behind it.
3. Never twice in a row — after a recap is shown, the next one waits for fresh activity.
4. Interactive mode only (ctx.hasUI); skipped in print/headless mode.

The practical effect (a recap waiting when you return) is the same; the trigger is time-based rather than focus-based.

Configuration

Auto-recap is on by default. Disable it with an env var:

export PI_ENABLE_AWAY_RECAP=0

/recap always works regardless of this setting.

Install

.pi/settings.json:

{
  "packages": ["npm:@arvoretech/pi-recap"]
}