Discovered during #2336 sweep (PR #2463). Not part of the issue AC — filing as dedicated follow-up.

Finding

The following exports in `src/auto-reply/reply/agent-runner-utils.ts` have zero production callers — dead Pi-embedded plumbing, only self-tested:

• `buildEmbeddedRunBaseParams` (line ~171)
• `buildEmbeddedContextFromTemplate` (line ~203)
• `buildEmbeddedRunContexts` (line ~249)
• `buildTemplateSenderContext`
• `resolveRunAuthProfile`
• `resolveEnforceFinalTag`
• `resolveProviderScopedAuthProfile`
• `resolveModelFallbackOptions`
• `buildThreadingToolContext`

Verification: `git grep -l "" -- 'src/**' | grep -v agent-runner-utils | grep -v "\.test\."` returns 0 hits for each.

Consumers of these via `FollowupRun.run.{verboseLevel, reasoningLevel, elevatedLevel, execOverrides, bashElevated}` are also dead — only the dead cluster reads these fields. The live usage of `verboseLevel` (tool-output gating in `agent-runner-helpers.ts`) reads directly from the session store, NOT from `FollowupRun.run`.

Scope

• Delete the 9 dead functions in `agent-runner-utils.ts`.
• Delete their self-tests in `agent-runner-utils.test.ts`.
• Remove `verboseLevel` / `reasoningLevel` / `elevatedLevel` / `execOverrides` / `bashElevated` from `FollowupRun.run` type in `src/auto-reply/reply/queue/types.ts`.
• Remove corresponding assignments in `get-reply-run.ts` (`verboseLevel: resolvedVerboseLevel`, etc.) — keep `resolvedVerboseLevel` local since it flows to agent-runner tool-output gate.
• Keep `verboseLevel` / `reasoningLevel` / `elevatedLevel` as session-store fields (load-bearing for display / security / ACP surfaces).
• Keep `resolveOriginMessageTo` and `resolveRunModelFallbacksOverride` (3 and 1 production consumers respectively).

Related follow-ups

• gut(auto-reply): finish thinking-level and model-selection sweep — follow-up to #2335 #2336 (Area 5) may overlap (directive-handling params). Sequence so one doesn't block the other.
• Live verboseLevel consumer: `src/auto-reply/reply/agent-runner-helpers.ts::createShouldEmitToolResult` / `createShouldEmitToolOutput` — must continue reading from session store.

AC

• All 9 dead functions + their self-tests deleted.
• `FollowupRun.run` trimmed of dead fields.
• `pnpm check` + `pnpm test` pass.
• No new stub-debt / type suppressions introduced.

Context

Discovered in PR #2463 while tracing `resolvedVerboseLevel` / `resolvedReasoningLevel` / `resolvedElevatedLevel` consumption — the answer is "only via a dead function cluster." Filing separately to keep PR #2463 scoped to thinking-specific removals.