Feature Request: Add optional "forgotten memory" detection for stale TODOs in daily notes #393

Use Case

ReMe handles long-term memory well (summary, decay, retrieval), but short-lived daily notes (YYYY-MM-DD.md style) can accumulate unresolved items that fall through the cracks.

Problem

In a QwenPaw agent workspace, memory/YYYY-MM-DD.md files serve as short-lived "daily notes" — quick context dumps, decisions, and pending items. When the agent is interrupted or a context window collapses mid-session, these notes often end up containing:

- TODO: add validation for edge cases in stock report → not started
- BLOCKED: need user confirmation for API key rotation

There's currently no lifecycle management for these notes. They just sit there, silently. The content is ephemeral (shouldn't survive ~7 days), but the TODO/BLOCKED markers are actionable signals that the agent should revisit.

Proposed Feature: forgotten-memory-check Hook

Add an optional stale TODO detector that runs as part of ReMe's post-chat sync:

def check_forgotten_memories(memory_dir, max_age_days=7):
    """Scan daily notes for unresolved markers older than max_age_days."""
    stale = []
    cutoff = datetime.now() - timedelta(days=max_age_days)
    for note in Path(memory_dir).glob("*.md"):
        if note.stat().st_mtime > cutoff:
            continue  # Not old enough
        content = note.read_text()
        markers = re.findall(r'(?i)(TODO|FIXME|BLOCKED|待办|待确认)', content)
        if markers:
            stale.append({
                "file": note.name,
                "days_old": (datetime.now() - note.stat().st_mtime).days,
                "markers": list(set(m.upper() for m in markers))
            })
    return stale

How It Would Help

Integration Point

This could be exposed as:

1. A pre-chat hook — agent checks forgotten memories before each conversation starts
2. A ReMe.forget() extension — include stale-marker filtering in the existing cleanup path
3. A config option in sync_policy — "stale_marker_days": 7

Relationship to Existing Features

This complements (not replaces):

• summary — summarizes conversations, doesn't flag stale action items
• decay — degrades memory scores based on time, but operates on long-term files not short-term notes
• fetch — retrieves context, doesn't detect what's missing/forgotten

The gap: there's no mechanism to surface "this daily note had an unresolved TODO from 10 days ago and it was never addressed."

Alternatives Considered

1. External cron job — works but bypasses ReMe's memory lifecycle; creates a parallel system
2. Agent self-check — agents can do this, but it requires the agent to remember to check (ironic)
3. ReMe-level hook — keeps lifecycle management inside the memory system where it belongs

Would love to hear thoughts on whether this fits ReMe's design philosophy, or if there's a better integration point I'm not seeing.