<system-reminder> wrapping fires on every tool-call iteration in follow-up messages, derailing model behavior

[Flowing8175]

Description

In prompt.ts (~line 1415), user message text parts in follow-up messages get wrapped in tags with "Please address this message and continue with your tasks" whenever step > 1 && lastFinished. This fires on every tool-call round.
Models interpret "Please address this message and continue with your tasks" as a system-level instruction to keep going, and start outputting "The system is asking me to continue" and wastes tokens. The in-place mutation of p.text also causes the wrapping to nest on subsequent iterations, compounding the issue.

Plugins

code-yeongyu/oh-my-openagent, different-ai/opencode-browser

OpenCode version

1.15.11

Steps to reproduce

1. Start a session, send any message, let it complete (sets lastFinished)
2. Send a follow-up message that requires multiple tool calls (e.g. "fix all lint errors" or "read these files then summarize")
3. Every tool calls prepends <system-reminder> ...
4. The model starts repeating "the system is asking me to continue"

First message in a fresh session never triggers this since lastFinished is null so the wrapping block is skipped.

Screenshot and/or share link

Operating System

Ubuntu 26.04

Terminal

Konsole (bundled with Kubuntu)

(sorry for initial mispaste with ai-generated issue draft)

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• runLoop fails to exit when client-generated messageID has clock skew ahead of server, causing infinite continuation with <system-reminder> wrap #28618: Describes the same root cause — step > 1 && lastFinished in prompt.ts triggering the <system-reminder> wrap and causing spurious extra LLM round-trips. That issue focuses on clock skew as the trigger, while this issue focuses on normal multi-tool-call sessions, but the wrapping/nesting mechanism and the affected code path are the same.
• Synthetic <system-reminder> leaks into user messages and gets treated as user-authored input #19405: Synthetic <system-reminder> leaking into user messages and being treated as user-authored input — related symptom where the tag content influences model behavior unintentionally.