claude-code - 💡(How to fix) Fix Model verbalizes acknowledgment of <system-reminder> task-tool prompt to user (e.g. "Task reminder noted, not applicable here")

Summary

When the harness injects the periodic task-tool <system-reminder> between turns, the model frequently writes a user-facing line acknowledging or dismissing it — e.g. "Task reminder noted, not relevant here", "(Task reminder noted, not applicable)". From the user's side this appears out of nowhere: there was no user message mentioning a reminder, so the acknowledgment reads as random or hallucinated context.

A CLAUDE.md directive instructing the model to never acknowledge system reminders in user-facing text does not reliably suppress this behavior.

Exact reminder text being acknowledged

The task tools haven't been used recently. If you're working on tasks that would benefit from tracking progress, consider using TaskCreate to add new tasks and TaskUpdate to update task status (set to in_progress when starting, completed when done). Also consider cleaning up the task list if it has become stale. Only use these if relevant to the current work. This is just a gentle reminder - ignore if not applicable.

Expected behavior

The reminder is an internal harness signal. The model should either:

1. Act on it silently (call the task tool if appropriate), or
2. Ignore it silently.

It should not produce user-facing text referencing the reminder, since the user has no visibility into the reminder itself.

Actual behavior

Model appends meta-commentary to its response, such as:

• "Task reminder noted, not relevant here."
• "(Task reminder noted — not applicable.)"
• "Noting the system reminder; nothing to track here."

The user sees these unprompted, with no context for what "reminder" refers to.

Why this isn't covered by existing instructions

The reminder body itself ends with "ignore if not applicable" — which the model interprets as license to say it's ignoring it, rather than to silently ignore it. A CLAUDE.md rule telling the model not to verbalize the reminder is in the user's project context, but the reminder text is injected closer to the turn and tends to win.

Adjacent / related issues (not duplicates)

• #59479 — UX feedback: TaskCreate reminder cadence + permission-block opacity
• #56867 — System reminders inserted after every tool call disrupt long-running multi-step work
• #46465 — Harness-emitted <system-reminder> uses "NEVER mention this reminder to the user" phrasing indistinguishable from prompt injection
• #60323 — TaskCreate reminder fires despite explicit "Do NOT use TaskCreate" directive in CLAUDE.md
• #41091 (closed) — System task-tool reminders degrade session quality by pushing execution over conversation

None of these cover the specific failure mode in this report: the model writing user-facing text about the reminder.

Suggested fix

Append explicit guidance to the reminder text along the lines of:

Do not mention this reminder in your user-facing response. Either call the task tool silently or proceed as if the reminder were not present.

Or, equivalently, give the model a clear rule that <system-reminder> content is internal and must never be the subject of user-facing text, even meta-commentary like "noted" or "ignoring".

Environment

• Claude Code 2.1.140
• Windows 11
• VS Code integrated terminal, PowerShell 7
• Reminder observed firing 3+ times in a single session (after /clear and between unrelated tool calls)