chore: make agent mistakes fail early

.github/workflows/agent-context-budget.yml

@@ -51,5 +51,5 @@ jobs:
             echo "Accepted by the agent-context-increase label."
             exit 0
           fi
-          echo "::error file=$BUDGET::Agent context grew compared to main. Cut the added text, or ask a maintainer to add the agent-context-increase label."
+          echo "::error file=$BUDGET::Agent context grew compared to main. Cut the added text and rerun the evals that cover it (evals/README.md), or ask a maintainer to add the agent-context-increase label."
           exit 1

AGENTS.md

@@ -98,7 +98,7 @@ An empty database is a bad test. Seed your worktree's `.t3` with a copy of real
 
 ## Verifying
 
-- Before you push, run `vp run check:changed`. It runs CI's fast checks for changed files: format, lint, knip, typecheck of affected packages and dependents, and tests beside changes. CI still runs the full test suites.
+- Before you push, run `vp run check:changed`. It runs CI's fast checks for changed files: format, lint, knip, typecheck of affected packages and dependents, tests beside changes, and the cross-file checks those miss. CI still runs the full test suites.
 - While you iterate, run `vp test run <files>`. For big files, add `-t "<name>"`.
 - Use `vp run typecheck`, not `npx tsc`; it caches per package. Both run tsc-rs (`typescript` is the `@tsc-rs/typescript` drop-in).
 - Run `node -e` snippets from `apps/server` or the owning package. `effect` does not resolve from the repo root.

apps/mobile/src/features/threads/PendingUserInputCard.tsx

@@ -26,9 +26,9 @@ import { useMarkdownPreviewStyles } from "../files/FileMarkdownPreview";
 import { useLiveThreadLinkLabels } from "../../state/entities";
 import {
   isPendingUserInputOptionSelected,
-  type PendingUserInput,
   type PendingUserInputDraftAnswer,
-} from "../../lib/threadActivity";
+} from "@t3tools/client-runtime/state/pending-user-input";
+import type { PendingUserInput } from "../../lib/threadActivity";
 
 export interface PendingUserInputCardProps {
   readonly environmentId: EnvironmentId;

apps/mobile/src/features/threads/ThreadDetailScreen.tsx

@@ -157,10 +157,10 @@ import { appAtomRegistry } from "../../state/atom-registry";
 import { threadStreamHaptics } from "./threadStreamHaptics";
 import { useAtomCommand } from "../../state/use-atom-command";
 import { useDelayedStatus } from "../../lib/useDelayedStatus";
+import type { PendingUserInputDraftAnswer } from "@t3tools/client-runtime/state/pending-user-input";
 import type {
   PendingApproval,
   PendingUserInput,
-  PendingUserInputDraftAnswer,
   ThreadFeedEntry,
   ThreadFeedLatestRun,
 } from "../../lib/threadActivity";

apps/mobile/src/lib/threadActivity.test.ts

@@ -31,10 +31,6 @@ import {
   type ThreadFeedCache,
   type ThreadFeedEntry,
   type ThreadFeedLatestRun,
-  togglePendingUserInputOptionSelection,
-  setPendingUserInputCustomAnswer,
-  isPendingUserInputOptionSelected,
-  buildPendingUserInputAnswers,
 } from "./threadActivity";
 
 const threadId = ThreadId.make("thread-1");
@@ -1368,17 +1364,6 @@ describe("retained v2 feed presentation", () => {
   });
 });
 
-const singleSelectQuestion = {
-  id: "runtime",
-  header: "Runtime",
-  question: "Which runtime should be used?",
-  options: [
-    { label: "Go", description: "One binary" },
-    { label: "Node.js", description: "Reuse TypeScript" },
-  ],
-  multiSelect: false,
-} as const;
-
 describe("subagent continued by a later turn", () => {
   const run2 = RunId.make("run-2");
   const at = (second: number) => `2026-10-09T10:00:${String(second).padStart(2, "0")}.000Z`;
@@ -1454,95 +1439,6 @@ describe("subagent continued by a later turn", () => {
   });
 });
 
-describe("pending user input answers", () => {
-  it("preserves exact editor text, including a deliberately cleared answer", () => {
-    const question = { ...singleSelectQuestion, initialAnswer: "  Proposed message\n" };
-    expect(
-      buildPendingUserInputAnswers([question], {
-        runtime: { customAnswer: question.initialAnswer },
-      }),
-    ).toEqual({ runtime: question.initialAnswer });
-    expect(
-      buildPendingUserInputAnswers([question], {
-        runtime: { customAnswer: "  Edited message\n\n" },
-      }),
-    ).toEqual({ runtime: "  Edited message\n\n" });
-    expect(buildPendingUserInputAnswers([question], { runtime: { customAnswer: "" } })).toEqual({
-      runtime: "",
-    });
-    expect(buildPendingUserInputAnswers([question], { runtime: { customAnswer: " \n" } })).toEqual({
-      runtime: " \n",
-    });
-    expect(setPendingUserInputCustomAnswer(question, { selectedOptionValues: ["Go"] }, "")).toEqual(
-      { customAnswer: "" },
-    );
-  });
-});
-
-describe("provider question values", () => {
-  const question = {
-    ...singleSelectQuestion,
-    allowCustomAnswer: false,
-    options: [
-      { label: "Same label", value: "  exact first  ", description: "First" },
-      { label: "Same label", value: "second", description: "Second" },
-    ],
-  } as const;
-
-  it("submits raw option values and distinguishes duplicate labels", () => {
-    const first = togglePendingUserInputOptionSelection(question, undefined, "  exact first  ");
-    expect(isPendingUserInputOptionSelected(question, first, "  exact first  ")).toBe(true);
-    expect(isPendingUserInputOptionSelected(question, first, "second")).toBe(false);
-    expect(buildPendingUserInputAnswers([question], { runtime: first })).toEqual({
-      runtime: "  exact first  ",
-    });
-    expect(togglePendingUserInputOptionSelection(question, first, "Same label")).toBe(first);
-  });
-
-  it("rejects arbitrary text when the provider only accepts offered options", () => {
-    expect(setPendingUserInputCustomAnswer(question, undefined, "Other")).toEqual({});
-    expect(
-      buildPendingUserInputAnswers([question], { runtime: { customAnswer: "Other" } }),
-    ).toBeNull();
-    expect(
-      buildPendingUserInputAnswers([question], { runtime: { selectedOptionValues: ["unknown"] } }),
-    ).toBeNull();
-    expect(
-      buildPendingUserInputAnswers([question], {
-        runtime: { selectedOptionValues: ["second"], customAnswer: "stale draft" },
-      }),
-    ).toEqual({ runtime: "second" });
-  });
-});
-
-it("accepts ready attachment-only answers while preserving selected options", () => {
-  const question = {
-    id: "q",
-    header: "Spec",
-    question: "Provide a specification",
-    options: [{ label: "Yes", description: "Approve" }],
-    multiSelect: false,
-  };
-  expect(buildPendingUserInputAnswers([question], { q: { attachmentCount: 1 } })).toEqual({
-    q: "",
-  });
-  expect(
-    buildPendingUserInputAnswers([question], {
-      q: { attachmentCount: 1, selectedOptionValues: ["Yes"] },
-    }),
-  ).toEqual({ q: "Yes" });
-  expect(
-    buildPendingUserInputAnswers([question], {
-      q: { attachmentCount: 1, attachmentsBlocked: true },
-    }),
-  ).toBeNull();
-  expect(
-    buildPendingUserInputAnswers([{ ...question, allowCustomAnswer: false }], {
-      q: { attachmentCount: 1 },
-    }),
-  ).toBeNull();
-});
-
 it("stops stranded thinking after a steer and follows the next thought or tool", () => {
   const at = "2026-06-20T00:00:02.000Z";
   const thought = (id: string): OrchestrationV2TurnItem => ({

apps/mobile/src/lib/threadActivity.ts

@@ -1,7 +1,6 @@
 import type {
   ThreadPendingApproval,
   ThreadPendingUserInput,
-  ThreadUserInputQuestion,
 } from "@t3tools/client-runtime/state/thread-requests";
 import { turnItemIsWorkspacePreparation } from "@t3tools/client-runtime/state/turn-item-presentation";
 import type { ThreadCheckpointSummary } from "@t3tools/client-runtime/state/thread-checkpoints";
@@ -81,13 +80,6 @@ import {
 export type PendingApproval = ThreadPendingApproval;
 export type PendingUserInput = ThreadPendingUserInput;
 
-export interface PendingUserInputDraftAnswer {
-  readonly selectedOptionValues?: ReadonlyArray<string>;
-  readonly customAnswer?: string;
-  readonly attachmentCount?: number;
-  readonly attachmentsBlocked?: boolean;
-}
-
 export interface ThreadFeedActivity {
   readonly id: string;
   readonly createdAt: string;
@@ -406,73 +398,6 @@ function isUserInputActivityGroup(entry: ThreadFeedActivityGroup): boolean {
   return entry.activities.some((activity) => activity.workEntry.questionAnswer !== undefined);
 }
 
-function normalizeDraftAnswer(value: string | undefined): string | null {
-  if (typeof value !== "string") return null;
-  const trimmed = value.trim();
-  return trimmed.length > 0 ? trimmed : null;
-}
-
-function resolvePendingUserInputOptionValue(
-  question: ThreadUserInputQuestion,
-  value: string,
-): string | null {
-  if (question.options.some((option) => option.value === value)) {
-    return value;
-  }
-
-  const label = value.trim();
-  return label.length > 0 &&
-    question.options.some((option) => option.value === undefined && option.label.trim() === label)
-    ? label
-    : null;
-}
-
-function normalizeSelectedOptionValues(
-  question: ThreadUserInputQuestion,
-  value: ReadonlyArray<string> | undefined,
-): ReadonlyArray<string> {
-  if (!Array.isArray(value)) {
-    return [];
-  }
-
-  return Array.from(
-    new Set(
-      value
-        .map((entry) => resolvePendingUserInputOptionValue(question, entry))
-        .filter((entry): entry is string => entry !== null),
-    ),
-  );
-}
-
-function resolvePendingUserInputAnswer(
-  question: ThreadUserInputQuestion,
-  draft: PendingUserInputDraftAnswer | undefined,
-): string | ReadonlyArray<string> | null {
-  if (draft?.attachmentsBlocked) return null;
-  const customAnswer =
-    question.allowCustomAnswer === false
-      ? null
-      : question.initialAnswer !== undefined
-        ? (draft?.customAnswer ?? null)
-        : normalizeDraftAnswer(draft?.customAnswer);
-  if (customAnswer !== null && customAnswer.length > 0) {
-    return customAnswer;
-  }
-
-  const selectedOptionValues = normalizeSelectedOptionValues(question, draft?.selectedOptionValues);
-  if (question.multiSelect) {
-    return selectedOptionValues.length > 0
-      ? selectedOptionValues
-      : (customAnswer ??
-          (question.allowCustomAnswer !== false && (draft?.attachmentCount ?? 0) > 0 ? "" : null));
-  }
-  return (
-    selectedOptionValues[0] ??
-    customAnswer ??
-    (question.allowCustomAnswer !== false && (draft?.attachmentCount ?? 0) > 0 ? "" : null)
-  );
-}
-
 function capitalizePhrase(value: string): string {
   const trimmed = value.trim();
   return trimmed.length === 0 ? value : `${trimmed.charAt(0).toUpperCase()}${trimmed.slice(1)}`;
@@ -1974,93 +1899,6 @@ function liveToolActivitySummary(activity: ThreadFeedActivity, presentTense: boo
   return activity.detail ?? activity.summary;
 }
 
-export function setPendingUserInputCustomAnswer(
-  question: ThreadUserInputQuestion,
-  draft: PendingUserInputDraftAnswer | undefined,
-  customAnswer: string,
-): PendingUserInputDraftAnswer {
-  if (question.allowCustomAnswer === false) {
-    return draft ?? {};
-  }
-
-  const selectedOptionValues =
-    question.initialAnswer !== undefined || customAnswer.trim().length > 0
-      ? undefined
-      : normalizeSelectedOptionValues(question, draft?.selectedOptionValues);
-  return {
-    customAnswer,
-    ...(selectedOptionValues && selectedOptionValues.length > 0 ? { selectedOptionValues } : {}),
-  };
-}
-
-export function isPendingUserInputOptionSelected(
-  question: ThreadUserInputQuestion,
-  draft: PendingUserInputDraftAnswer | undefined,
-  optionValue: string,
-): boolean {
-  if (question.allowCustomAnswer !== false && normalizeDraftAnswer(draft?.customAnswer)) {
-    return false;
-  }
-
-  const resolvedOptionValue = resolvePendingUserInputOptionValue(question, optionValue);
-  return (
-    resolvedOptionValue !== null &&
-    normalizeSelectedOptionValues(question, draft?.selectedOptionValues).includes(
-      resolvedOptionValue,
-    )
-  );
-}
-
-export function togglePendingUserInputOptionSelection(
-  question: ThreadUserInputQuestion,
-  draft: PendingUserInputDraftAnswer | undefined,
-  optionValue: string,
-): PendingUserInputDraftAnswer {
-  const resolvedOptionValue = resolvePendingUserInputOptionValue(question, optionValue);
-  if (resolvedOptionValue === null) {
-    return draft ?? {};
-  }
-
-  if (question.multiSelect) {
-    const selectedOptionValues = normalizeSelectedOptionValues(
-      question,
-      draft?.selectedOptionValues,
-    );
-    const nextSelectedOptionValues = selectedOptionValues.includes(resolvedOptionValue)
-      ? selectedOptionValues.filter((value) => value !== resolvedOptionValue)
-      : [...selectedOptionValues, resolvedOptionValue];
-
-    return {
-      customAnswer: "",
-      ...(nextSelectedOptionValues.length > 0
-        ? { selectedOptionValues: nextSelectedOptionValues }
-        : {}),
-    };
-  }
-
-  return {
-    customAnswer: "",
-    selectedOptionValues: [resolvedOptionValue],
-  };
-}
-
-export function buildPendingUserInputAnswers(
-  questions: ReadonlyArray<ThreadUserInputQuestion>,
-  draftAnswers: Record<string, PendingUserInputDraftAnswer>,
-): Record<string, string | ReadonlyArray<string>> | null {
-  const answers: Record<string, string | ReadonlyArray<string>> = {};
-
-  for (const question of questions) {
-    const answer = resolvePendingUserInputAnswer(question, draftAnswers[question.id]);
-    if (answer === null) {
-      return null;
-    }
-    answers[question.id] = answer;
-  }
-
-  return answers;
-}
-
 interface ThreadFeedBuild {
   readonly rows: ReadonlyArray<OrchestrationV2ProjectedTurnItem>;
   readonly attempts: ReadonlyArray<OrchestrationV2RunAttempt> | undefined;

apps/mobile/src/state/use-selected-thread-requests.ts

@@ -42,7 +42,7 @@ import {
   setPendingUserInputCustomAnswer,
   togglePendingUserInputOptionSelection,
   type PendingUserInputDraftAnswer,
-} from "../lib/threadActivity";
+} from "@t3tools/client-runtime/state/pending-user-input";
 import { appAtomRegistry } from "./atom-registry";
 import { useSelectedThreadPendingRequests } from "./use-thread-detail";
 import { useThreadSelection } from "./use-thread-selection";

apps/server/src/provider/AgentContextBudget.test.ts

@@ -123,7 +123,10 @@ it.effect("agent context matches the committed budget", () =>
       ),
     };
 
-    expect(measured).toEqual(budget);
+    expect(
+      measured,
+      "Set agentContextBudget.json to the measured values. Before you cut text to fit, run the evals that cover it (evals/README.md).",
+    ).toEqual(budget);
   }).pipe(
     Effect.provide(
       McpHttpServer.layerTools.pipe(

apps/web/src/components/chat/ChatComposer.tsx

@@ -241,7 +241,7 @@ import {
   type ThreadShell,
 } from "../../types";
 import { type ComposerPromptHistoryPosition } from "./composerPromptHistory";
-import type { PendingUserInputDraftAnswer } from "../../pendingUserInput";
+import type { PendingUserInputDraftAnswer } from "@t3tools/client-runtime/state/pending-user-input";
 import type {
   LatestProposedPlanState,
   PendingApproval,

apps/web/src/components/chat/ComposerPendingUserInputPanel.tsx

@@ -3,10 +3,8 @@ import { memo, use, useEffect, useEffectEvent, useRef, useState } from "react";
 import { BackgroundThreadContext } from "./backgroundThread";
 import { useStableCallback } from "~/hooks/useStableCallback";
 import { type PendingUserInput } from "../../session-logic";
-import {
-  derivePendingUserInputProgress,
-  type PendingUserInputDraftAnswer,
-} from "../../pendingUserInput";
+import type { PendingUserInputDraftAnswer } from "@t3tools/client-runtime/state/pending-user-input";
+import { derivePendingUserInputProgress } from "../../pendingUserInput";
 import { CheckboxMark } from "../ui/checkbox";
 import { Collapsible, CollapsiblePanel, CollapsibleTrigger } from "../ui/collapsible";
 import { Kbd } from "../ui/kbd";