Problem

/explain and /optimize output is too verbose for terminal use. The AI response streams full prose paragraphs that are hard to scan quickly in a terminal session.

Goal

Make both commands feel like a senior DBA giving you a quick answer at the terminal, not a blog post. Compact, high-signal, immediately actionable.

Proposed output format

/explain

-- EXPLAIN ANALYZE output (shown as today) --

[AI]
Bottleneck: Seq Scan on orders (cost=0.00..4821.00, 98K rows) — consider index on status
Est. error: 12x overestimate on nested loop (est 1 vs actual 12)
Time: planning 2ms, execution 847ms

Rules for AI section:

• Max 4 lines, no headers, no markdown bullet lists
• Lead with the single most expensive node
• One line per finding (bottleneck / row-estimate error / wait correlation if present)
• End with planning + execution time summary
• No prose, no "In this query...", no restating what the user can already see

/optimize

-- EXPLAIN ANALYZE output --

[AI]
1. CREATE INDEX orders_status_idx ON orders(status); -- seq scan → index scan, est. 10x speedup
2. Rewrite: replace correlated subquery in line 3 with LEFT JOIN
3. Vacuum orders — 40K dead tuples, last vacuum 14 days ago

Rules for AI section:

• Ordered list of concrete actions, highest impact first
• Each line: action + single-line rationale
• CREATE INDEX suggestions must be copy-pasteable SQL
• No prose intro, no "Based on the plan...", no section headers

Implementation

• Modify system prompts in handle_ai_explain and handle_ai_optimize in src/repl/ai_commands.rs
• Update max_tokens budget for explain/optimize (current default likely too high for compact output — suggest 300-400 tokens)
• Add unit tests for the prompt structure
• Test with real queries against local Postgres and paste evidence in PR

Acceptance criteria

• /explain AI section fits in ≤4 terminal lines for a typical query
• /optimize returns ≤5 numbered items, each ≤1 line
• No change to the raw EXPLAIN ANALYZE output shown before the AI section
• Existing tests pass