what-was-i-thinking — L2 monthly snapshot

Framework context

L2 skill bounded to a single YYYY-MM window. Computes monthly averages, classifies activity as HIGH (>1.5× monthly avg) / NORMAL / LOW (<0.5×). Unlike tunnel-state, this is a temporal window not a topical window.

When to invoke

• User cites a specific month: "what was I doing in 2025-10"
• Reviewing a past hyperfocus period
• Before what-do-i-think, to orient temporally

Input

month: str   # YYYY-MM format, strict

Output contract

## 📅 What you were thinking: {month}
**Activity level**: 🔥 HIGH / 📊 NORMAL / 📉 LOW (N msgs vs avg M)
**Conversations**: N · **User messages**: M · **Questions asked**: K

### Top conversations (by message count)
1. **{title}** ({msg_count} messages) [conv_id · date]
...

### Sample questions that month
- "{q truncated 150ch}" [conv_id · date]
...

Does NOT do

• Accept relative dates ("last month") — caller resolves to YYYY-MM
• Synthesize what the month "meant" — only structural stats + verbatim samples
• Compare months (that's trajectory)

Verification checklist

• [ ] Month format strictly YYYY-MM (return error on malformed input)
• [ ] Activity level uses the 1.5× / 0.5× thresholds documented
• [ ] Top conversations cite conv_id
• [ ] Sample questions are verbatim (truncated 150ch)