Resumed coding-agent harness session returns the previous turn's answer verbatim as a successful reply to a new prompt

In the AI SDK OpenCode harness, when a resumed session's event stream closed before the new turn settled, a context fallback emitted the session's newest persisted assistant message plus a synthetic successful finish-step; on a cold resume that message was the previous turn's answer. The reporter hit this three times in production, with the caller seeing a normal completion. The merged fix fences the fallback against the pre-prompt assistant and makes interrupted turns fail instead of synthesizing success.

Framework / agent

Vercel AI SDK (@ai-sdk/harness-opencode) · OpenCode coding agent run through the AI SDK harness with persistent sandboxes

Remediation attempts

TESTEDSUGGESTED

Recurrence

not documented

Source languages

en

Updated

2026-09-30

Sources

• GITHUB ISSUE harness-opencode: cold resume can emit the PREVIOUS turn's assistant message as the reply (uncorrelated emitContextFallback) — github.com/vercel/ai, retrieved 2026-09-29
• GITHUB PULL REQUEST fix: prevent stale previous-turn replies on interrupted OpenCode session resumes — github.com/vercel/ai, retrieved 2026-09-29
• GITHUB PULL REQUEST fix(harness-opencode): don't emit a pre-turn assistant message as the turn's reply — github.com/vercel/ai, retrieved 2026-09-29

Symptoms

• The user receives the previous turn's answer verbatim, without its tool calls, as the reply to a new question
• No error surfaces and the turn is recorded as a normal successful completion

Similarity to your system is not implied. A remediation that worked in the documented context may not work in yours.