diff --git a/AGENTS.md b/AGENTS.md
index 98e0636b18..5035905f71 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -18,7 +18,7 @@
 ## Unit Tests Guidelines
 - New unit tests should be added to the `tests` module in the corresponding dialect file (e.g., `tests/sqlparser_redshift.rs` for Redshift), and should be placed at the end of the file.
 - If the new functionality is gated using a dialect function, and the SQL is likely relevant in most dialects, tests should be placed under `tests/sqlparser_common.rs`.
-- Cover both positive and negative cases: valid input parses and round-trips, invalid input fails with the expected error.
+- Cover both positive and negative cases: valid input parses and round-trips, invalid input fails with the expected error, asserted in full.
 - Cover every dialect the syntax is relevant to, e.g. via `all_dialects_where(...)`.
 - When adding new syntax, run the fuzzer to check for panics (see `docs/fuzzing.md`).
 - When fixing a bug, add one seed file to `fuzz/fuzz_seeds/`, as small as still triggers the original path. Name it `<dialect>_<issue>_<topic>.sql`. For a fix with no issue, use the PR number.
@@ -34,6 +34,15 @@ snowflake().one_statement_parses_to(
     "SELECT * FROM my_table AS t",
 )
 ```
+For invalid input, assert the exact error:
+```rust
+assert_eq!(
+    ParserError::ParserError("Expected: (, found: FROM".to_string()),
+    sqlite()
+        .parse_sql_statements("SELECT id, name FROM orders WHERE customer_id IN FROM blocked_customers")
+        .unwrap_err()
+);
+```
 
 ## Analyzing Parsing Issues
 You can try to simplify the SQL statement to identify the root cause of the parsing issue. This may involve removing certain clauses or components of the SQL statement to see if it can be parsed successfully. Additionally, you can compare the problematic SQL statement with similar statements that are parsed correctly to identify any differences that may be causing the issue.