feat(pi): drain and resume pi() runs

[eersnington]

pi() gets the same deploy behavior as piDurable(): live runs drain, and a run cut off by a stop resumes on wake when Pi can continue it.

	0.1.0	This PR
Actor stops mid-run	session.abort(), the run ends as aborted	waits for the run to end, then closes, inside sleepGracePeriod (15 min default)
Run outlives the drain	aborted	aborted in memory with nothing aborted stored, and resumes on the next wake if Pi accepts the transcript
Generation lost (the engine moved the actor)	aborted through c.abortSignal	aborted in memory at once, nothing stored; the next generation resumes it
prompt times out, or its caller leaves	run aborted	prompt rejects, and the run keeps going. Only the abort action stops it
Destroy	abort	abort

Drain and lost generations

• The drain deadline is taken before the app's own onSleep and keeps time back for closing (drainDeadline in src/drain.ts), so the final save and the sandbox suspend run before RivetKit's grace deadline.
• A lost generation skips onSleep, so stopRunWhenLost listens on the actor's abort signal from onWake and aborts the run when the context reports isLost. Otherwise the old process keeps calling the model and running tools next to the generation that resumed the run. RivetKit releases without lost generations still run onSleep, and the drain handles them.

When a run resumes

Resume never writes an entry Pi would not write. It continues only when Pi's own Agent.continue() accepts the stored transcript: the last message is a user or tool-result message.

Stopped while	Transcript ends with	Resumes?
the model was answering	the user message, or the last tool results	yes. The model is asked again, and the partial answer is lost
parallel tools ran and at least one finished	a tool result	yes. Unfinished calls get pi-ai's "No result provided"
a tool ran and none finished	the assistant message with the tool calls	no
Pi's own abort ran	an aborted assistant message	no, which is what an abort means

A chat that does not resume is left as after an abort, and the next message continues it.

How

• Migration 2 adds run_state and queued_json to pi_session. A run start records running and the queued steer and follow-up messages; the end records idle.
• On wake, after the app's onWake, resumeInterruptedRun replays the queue and drives AgentSession's own run loop, so retry, compaction, and queued messages work as in any run:

for (const text of queue.steering) await internals._queueSteer(text);
for (const text of queue.followUp) await internals._queueFollowUp(text);
// The loop starts with agent.prompt; for this one call it continues instead.
Object.defineProperty(agent, "prompt", { configurable: true, value: async () => {
	Reflect.deleteProperty(agent, "prompt");
	await agent.continue();
} });
await internals._runAgentPrompt([]);

• These are private methods. pi-coding-agent is pinned exactly, and the resume test fails if an upgrade changes them.
• A resumed run runs in the background under c.keepAwake. A client that was waiting on prompt when the actor stopped got an error; it sees the finished run through events or getMessages().

---

This is part 8 of 11 in a stack:

•  11  fix(pi): persist migrated pi() session entries #330
•  10  feat(pi): expose the pi session to actor code as c.pi #329
•  9  perf(pi): connect the pi() sandbox on first use #328
•  8  feat(pi): drain and resume pi() runs #327 👈
•  7  chore(examples): bump the docs examples to pi-coding-agent 1.0.2 #326
•  6  test(pi): let a run finish and the actor sleep when its caller disconnects #325
•  5  feat(pi): drain pi durable actors across upgrades #324
•  4  feat(pi): stream pi durable conversations to clients #323
•  3  feat(pi): add piDurable actor #322
•  2  feat(pi): store pi durable sessions in actor sqlite #321
•  1  feat(pi): add piDurable() and drain, resume, and c.pi for pi() #320