/reply

User shows you a thread. You return three plausible replies and a recommendation. They pick.

Recipe

1. Read the thread. Last 10–20 messages.

2. Identify the actual question or ask. Strip pleasantries. What does this message actually want from the user?

3. Read memory/people.md and memory/commitments.md for relevant context (e.g., have they already promised something here?).

4. Draft three options that span the strategy space:

   • Option A — Direct yes/agreement: shortest path forward.
   • Option B — Push back or qualify: where the user has constraints, hesitation, or a different read.
   • Option C — Buy time or redirect: defer, ask a clarifying question, change the venue.

5. Recommend one based on:

   • User's stated priorities in life.md
   • The relationship's pattern (from people.md)
   • Open commitments that might be affected

6. Format the output as:

**They said:** <one-line restatement of the ask>

**Option A — <flavor>**
<draft text>

**Option B — <flavor>**
<draft text>

**Option C — <flavor>**
<draft text>

**Recommendation: <A/B/C>.** <One sentence why.>

**Wildcard:** <one contrarian take they might not have considered>

7. Wait for user to pick. "Send A" → invoke /draft to send Option A.

Voice rules

Match the recipient's tone. See /draft for full voice rules.

When NOT to give 3 options

If the message clearly has only one reasonable response (e.g., a simple yes/no logistical question), just give the one reply and a note about why options aren't useful here.