feat: abort sub-agent jobs (real kill) + runner usage capture into JobRecord.usage
Wave 4 (agent-side observability + control), reconciled with the dock contract (master relay 2026-09): - abortJob in src/server/jobs.ts, exported through the server barrel for POST /api/activity/abort (master's thin route): fires the job's AbortController (real kill — the in-flight model call aborts, not just the next loop check), drops any queued result steer, flips the record to error "Aborted by learner" + activity "aborted" synchronously, and queues a one-line abort steer (buildAbortSteer) through the same wake pipeline so the teacher's promise doesn't hang in the lesson. Shares the result steer's deterministic id space, so abort + racing result can never both deliver (steerDelivered dedupe). - Abort registry (src/agent/subagents/aborts.ts) on globalThis, mirroring the wake pipeline's module-state pattern; runner registers/unregisters per run, so the route's module instance sees the loop's controller. - Runner usage capture: LoopContext (abortSignal + onUsage) threads into all three kind loops; per-step usage folds via accumulateJobUsage into JobRecord.usage (model from config MODELS, absent-never-zero), flushed in runJob's finally so an aborted job's money record survives the kill. - Contracts: JobUsage token fields optional + lastStep (most recent call's accounting, for the dock's per-blob context line vs CONTEXT_LIMITS); SubagentResultMeta.aborted; AbortSteerMessage; buildAbortSteer; SteerMessage union generalized. - demo-builder/visual-maker accept abortSignal + onUsage (visual folds generateText usage per iteration). - stillRunning/finishJob/jobContext guards: an aborted job is never overwritten by the runner (terminal no matter what the wake does); failJob and the pump's finalizeJob keep the flip. - UI: abort steer renders a quiet aborted chip (job-aborted-badge). - Prompt 20-jobs.md: one line about aborted jobs. Tests: accumulateJobUsage folds (undefined-preservation, lastStep), runner usage capture, real-kill abort (sync flip, silent unwind, abort steer wakes), usage-so-far survives the kill, aborting a delivering job drops the queued result steer, outcome reasons, abort-chip render. 330/330, typecheck/lint/build green.
This commit is contained in:
amosblomqvist
@@ -7,6 +7,7 @@
|
||||
* of "ai" (safe everywhere: server, client, tests). See docs/vision.md.
|
||||
*/
|
||||
import type { UIMessage } from "ai";
|
||||
import type { CallUsage } from "./chat.ts";
|
||||
|
||||
// ─── Async sub-agent jobs (vision §2.5; plan.md Wave 0 protocol) ────────────
|
||||
|
||||
@@ -36,18 +37,64 @@ export interface JobRecord {
|
||||
|
||||
/** A sub-agent run's accumulated usage — the same accounting shape as a
|
||||
* chat turn (contracts/chat.ts CallUsage), flattened to one running total
|
||||
* plus a step count. */
|
||||
* plus a step count. Token fields are optional exactly like CallUsage: a
|
||||
* field NO model call reported stays undefined — absent is not zero (the
|
||||
* money-channel doctrine; cost renders "—", never $0). */
|
||||
export interface JobUsage {
|
||||
/** The OpenRouter model id that ran the sub-agent. */
|
||||
model: string;
|
||||
/** Running totals summed over the run's model calls so far. */
|
||||
inputTokens: number;
|
||||
outputTokens: number;
|
||||
cacheReadTokens: number;
|
||||
cacheWriteTokens: number;
|
||||
reasoningTokens: number;
|
||||
inputTokens: number | undefined;
|
||||
outputTokens: number | undefined;
|
||||
cacheReadTokens: number | undefined;
|
||||
cacheWriteTokens: number | undefined;
|
||||
reasoningTokens: number | undefined;
|
||||
/** How many model calls the run has made (each bills separately). */
|
||||
steps: number;
|
||||
/** The MOST RECENT model call's accounting (2026-09 dock contract): the
|
||||
* per-blob context line reads last-step input+output against the model's
|
||||
* CONTEXT_LIMITS — never the cumulative totals. Absent until the first
|
||||
* call completes; absent provider fields stay absent (never zero). */
|
||||
lastStep?: CallUsage;
|
||||
}
|
||||
|
||||
/** The per-model-call usage shape we fold into JobUsage — the SDK's
|
||||
* LanguageModelUsage narrowed structurally (no runtime "ai" import: any
|
||||
* object with these fields folds, including the real thing and test fakes). */
|
||||
export interface UsageStepLike {
|
||||
inputTokens?: number | undefined;
|
||||
inputTokenDetails?: { cacheReadTokens?: number; cacheWriteTokens?: number } | undefined;
|
||||
outputTokens?: number | undefined;
|
||||
outputTokenDetails?: { reasoningTokens?: number } | undefined;
|
||||
}
|
||||
|
||||
/** Fold one model call's usage into the run's running total (pure). A
|
||||
* field neither the total nor the step reported stays undefined; once
|
||||
* reported, later unreported calls contribute 0. The `steps` counter is
|
||||
* the billing count: one per model call, reported or not. */
|
||||
export function accumulateJobUsage(
|
||||
total: JobUsage | undefined,
|
||||
model: string,
|
||||
step: UsageStepLike,
|
||||
): JobUsage {
|
||||
const add = (a: number | undefined, b: number | undefined): number | undefined =>
|
||||
a === undefined && b === undefined ? undefined : (a ?? 0) + (b ?? 0);
|
||||
return {
|
||||
model,
|
||||
inputTokens: add(total?.inputTokens, step.inputTokens),
|
||||
outputTokens: add(total?.outputTokens, step.outputTokens),
|
||||
cacheReadTokens: add(total?.cacheReadTokens, step.inputTokenDetails?.cacheReadTokens),
|
||||
cacheWriteTokens: add(total?.cacheWriteTokens, step.inputTokenDetails?.cacheWriteTokens),
|
||||
reasoningTokens: add(total?.reasoningTokens, step.outputTokenDetails?.reasoningTokens),
|
||||
steps: (total?.steps ?? 0) + 1,
|
||||
lastStep: {
|
||||
inputTokens: step.inputTokens,
|
||||
outputTokens: step.outputTokens,
|
||||
cacheReadTokens: step.inputTokenDetails?.cacheReadTokens,
|
||||
cacheWriteTokens: step.inputTokenDetails?.cacheWriteTokens,
|
||||
reasoningTokens: step.outputTokenDetails?.reasoningTokens,
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -155,6 +202,11 @@ export interface SubagentResultMeta {
|
||||
jobId: string;
|
||||
kind: JobKind;
|
||||
toolCallId: string;
|
||||
/** Present only on an abort steer (2026-09 abort semantics): the
|
||||
* learner aborted the job (it lands as error "Aborted by learner");
|
||||
* the teacher is told so its promise of a result doesn't hang
|
||||
* silently in the lesson. */
|
||||
aborted?: true;
|
||||
}
|
||||
|
||||
/** The data-subagent-artifact part payload: what the UI renders. */
|
||||
@@ -178,6 +230,20 @@ export interface SubagentSteerMessage extends UIMessage {
|
||||
];
|
||||
}
|
||||
|
||||
/** The abort steer: user-role like any steer, but NO artifact part —
|
||||
* there is no result to render, only the teacher-facing text. Shares the
|
||||
* result steer's deterministic id space (`steer_${jobId}`): steerDelivered
|
||||
* matches on metadata jobId, so an abort and a (racing) result steer
|
||||
* can never BOTH deliver — exactly one wakes the teacher. */
|
||||
export interface AbortSteerMessage extends UIMessage {
|
||||
role: "user";
|
||||
metadata: { subagentResult: SubagentResultMeta & { aborted: true } };
|
||||
parts: [{ type: "text"; text: string }];
|
||||
}
|
||||
|
||||
/** Anything the wake pipeline may queue and deliver. */
|
||||
export type SteerMessage = SubagentSteerMessage | AbortSteerMessage;
|
||||
|
||||
/** Whether a result is the failure shape (html/svg empty, failure brief). */
|
||||
export function isSubagentFailure(kind: JobKind, result: ResearchResult | DemoResult | VisualResult): boolean {
|
||||
if (kind === "research") {
|
||||
@@ -234,6 +300,32 @@ export function buildSubagentSteer(
|
||||
};
|
||||
}
|
||||
|
||||
/** Build the abort steer for an aborted job (pure — shared by the
|
||||
* abort path in src/server/jobs.ts and tests). Same id space as the result
|
||||
* steer; see AbortSteerMessage. */
|
||||
export function buildAbortSteer(
|
||||
job: Pick<JobRecord, "id" | "kind" | "toolCallId">,
|
||||
): AbortSteerMessage {
|
||||
return {
|
||||
id: `steer_${job.id}`,
|
||||
role: "user",
|
||||
metadata: {
|
||||
subagentResult: { jobId: job.id, kind: job.kind, toolCallId: job.toolCallId, aborted: true },
|
||||
},
|
||||
parts: [{ type: "text", text: abortText(job.kind) }],
|
||||
};
|
||||
}
|
||||
|
||||
/** What the teacher reads: one line telling it the result is not coming,
|
||||
* how to acknowledge, and that re-dispatching is allowed. */
|
||||
function abortText(kind: JobKind): string {
|
||||
if (kind === "research")
|
||||
return "Research aborted — the learner aborted the research job, so no brief is coming. Acknowledge it in one line and continue teaching without it (or offer to re-dispatch it).";
|
||||
if (kind === "demo")
|
||||
return "The demo job was aborted by the learner — no demo is coming. Acknowledge it in one line and teach the topic in prose instead (a re-dispatch is fine if the learner wants it).";
|
||||
return "The figure job was aborted by the learner — no figure is coming. Acknowledge it in one line and teach without it (a re-dispatch is fine if the learner wants it).";
|
||||
}
|
||||
|
||||
/** True when the transcript already carries this job's steer (idempotent delivery). */
|
||||
export function steerDelivered(messages: UIMessage[], jobId: string): boolean {
|
||||
return messages.some(
|
||||
|
||||
Reference in New Issue
Block a user