chore(ph-ai): adjusted sql prompt guardralsi (#42276)

Author: tatoalo

ee/hogai/chat_agent/sql/prompts.py

@@ -2,6 +2,16 @@
 You are an expert in writing HogQL. HogQL is PostHog's variant of SQL that supports most of ClickHouse SQL. We're going to use terms "HogQL" and "SQL" interchangeably.
 You write HogQL based on a prompt. You don't help with other knowledge. You are provided with the current HogQL query that the user is editing. You have access to the core memory about the user's company and product in the <core_memory> tag. Use this memory in your responses.
 
+CRITICAL - Function name casing:
+- HogQL function names are CASE-SENSITIVE and use camelCase (not snake_case or lowercase).
+- Common mistakes to avoid:
+  - WRONG: format_datetime, formatdatetime → CORRECT: formatDateTime
+  - WRONG: to_timezone, totimezone → CORRECT: toTimeZone
+  - WRONG: todatetime, to_datetime → CORRECT: toDateTime
+  - WRONG: to_date, todate → CORRECT: toDate
+  - WRONG: countif → CORRECT: countIf
+- Timezone strings are also case-sensitive: use 'UTC' not 'utc', 'America/New_York' not 'america/new_york'.
+
 Important HogQL differences versus other SQL dialects:
 - JSON properties are accessed using `properties.foo.bar` instead of `properties->foo->bar` for property keys without special characters.
 - JSON properties can also be accessed using `properties.foo['bar']` if there's any special character (note the single quotes).
@@ -181,7 +191,7 @@
 ## Functions and aggregations
 
 You can filter, modify, or aggregate accessed data with [supported ClickHouse functions](/docs/sql/clickhouse-functions.md) like `dateDiff()` and `concat()` and [aggregations](/docs/hogql/aggregations.md) like `sumIf()` and `count()`.
-Functions are always written in camel case for example `countIf()` instead of `COUNTIF()` or `COUNTIf`.
+REMINDER: Functions use camelCase (e.g., `countIf`, `formatDateTime`, `toTimeZone`) - see the CRITICAL section above.
 
 Here are some of the most common and useful ones:
 

ee/hogai/eval/ci/max_tools/eval_hogql_generator_tool.py

@@ -234,6 +234,13 @@ async def eval_tool_generate_hogql_query(call_generate_hogql_query, database_sch
                 expected="SELECT p.id AS person_id, p.properties.email AS email, p.properties.org AS organization FROM persons p WHERE p.properties.email NOT LIKE '%@test.com' AND (coalesce(variables.org, '') = '' OR p.properties.org = variables.org) ORDER BY p.created_at DESC LIMIT 1000",
                 metadata=metadata,
             ),
+            EvalCase(
+                input=EvalInput(
+                    instructions="Show me daily event counts for 'buy_button_clicked' events from Nov 25-26 2025, formatted as YYYY-MM-DD in UTC timezone",
+                ),
+                expected="SELECT formatDateTime(toTimeZone(timestamp, 'UTC'), '%Y-%m-%d') AS day, count() FROM events WHERE event = 'buy_button_clicked' AND timestamp >= toDateTime('2025-11-25 00:00:00') AND timestamp <= toDateTime('2025-11-26 23:59:59') GROUP BY day ORDER BY day DESC",
+                metadata=metadata,
+            ),
         ],
         pytestconfig=pytestconfig,
     )