From 26e88ce2a024f981f169db99737566b240e8e995 Mon Sep 17 00:00:00 2001
From: Mohamed Firas <62911996+Fir121@users.noreply.github.com>
Date: Wed, 18 Mar 2026 12:32:46 +0400
Subject: [PATCH] fix bug with braces in string during format

---
 pyproject.toml                     |  2 +-
 src/llm_classifier/classifier.py   |  8 ++++++--
 src/llm_classifier/cluster.py      |  7 +++++--
 src/llm_classifier/prompt_utils.py | 17 +++++++++++++++++
 tests/test_classifier.py           | 12 ++++++++++++
 tests/test_cluster.py              | 21 +++++++++++++++++++++
 6 files changed, 62 insertions(+), 5 deletions(-)
 create mode 100644 src/llm_classifier/prompt_utils.py

diff --git a/pyproject.toml b/pyproject.toml
index 6339278..2cb8fb8 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "llm-classifier"
-version = "0.1.2"
+version = "0.1.3"
 description = "Structured LLM based classification, clustering and extraction framework that works with all major API providers"
 readme = "README.md"
 license = { file = "LICENSE" }
diff --git a/src/llm_classifier/classifier.py b/src/llm_classifier/classifier.py
index ad54865..2651020 100644
--- a/src/llm_classifier/classifier.py
+++ b/src/llm_classifier/classifier.py
@@ -14,6 +14,8 @@
 import instructor
 from pydantic import BaseModel, Field, create_model
 
+from .prompt_utils import safe_prompt_format
+
 T = TypeVar("T", bound=BaseModel)
 
 
@@ -299,7 +301,8 @@ def _build_prompt(
         if not system_prompt:
             system = None
         else:
-            system = system_prompt.format(
+            system = safe_prompt_format(
+                system_prompt,
                 examples=examples_str,
                 format=format_schema,
                 input=input_text,
@@ -309,7 +312,8 @@ def _build_prompt(
         if not user_prompt:
             user = None
         else:
-            user = user_prompt.format(
+            user = safe_prompt_format(
+                user_prompt,
                 examples=examples_str,
                 format=format_schema,
                 input=input_text,
diff --git a/src/llm_classifier/cluster.py b/src/llm_classifier/cluster.py
index 9954494..80b594b 100644
--- a/src/llm_classifier/cluster.py
+++ b/src/llm_classifier/cluster.py
@@ -9,6 +9,8 @@
 import instructor
 from pydantic import BaseModel, Field, create_model
 
+from .prompt_utils import safe_prompt_format
+
 T = TypeVar("T", bound=BaseModel)
 
 
@@ -288,7 +290,8 @@ def _build_prompt(
         if not system_prompt:
             system = None
         else:
-            system = system_prompt.format(
+            system = safe_prompt_format(
+                system_prompt,
                 format=format_schema,
                 n_clusters_instruction=n_clusters_instruction,
                 validation_rules=validation_rules,
@@ -298,7 +301,7 @@ def _build_prompt(
         if not user_prompt:
             user = None
         else:
-            user = user_prompt.format(items=items_str)
+            user = safe_prompt_format(user_prompt, items=items_str)
 
         return system, user
 
diff --git a/src/llm_classifier/prompt_utils.py b/src/llm_classifier/prompt_utils.py
new file mode 100644
index 0000000..8263419
--- /dev/null
+++ b/src/llm_classifier/prompt_utils.py
@@ -0,0 +1,17 @@
+"""Prompt templating helpers."""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+def safe_prompt_format(template: str, **values: Any) -> str:
+    """Replace known placeholders without parsing other braces.
+
+    Unlike ``str.format``, this helper only replaces exact ``{name}`` tokens for
+    provided values and leaves all other braces untouched.
+    """
+    rendered = template
+    for key, value in values.items():
+        rendered = rendered.replace(f"{{{key}}}", str(value))
+    return rendered
diff --git a/tests/test_classifier.py b/tests/test_classifier.py
index 233685a..17713dc 100644
--- a/tests/test_classifier.py
+++ b/tests/test_classifier.py
@@ -159,6 +159,18 @@ def test_build_prompt_includes_examples(self, mock_classifier):
         assert "Great!" in system
         assert "Examples" in system
 
+    def test_build_prompt_allows_literal_braces(self, mock_classifier):
+        """Prompt templates with literal braces should not raise formatting errors."""
+        system, user = mock_classifier._build_prompt(
+            "payload {with} braces",
+            Sentiment,
+            system_prompt="Schema: {format} | keep literal {json}",
+            user_prompt="Input: {input} | keep literal {meta}",
+        )
+        assert "keep literal {json}" in system
+        assert "keep literal {meta}" in user
+        assert "payload {with} braces" in user
+
 
 # ============================================================================
 # Result Types Tests
diff --git a/tests/test_cluster.py b/tests/test_cluster.py
index 0b87cd0..e4b7360 100644
--- a/tests/test_cluster.py
+++ b/tests/test_cluster.py
@@ -207,6 +207,27 @@ def test_build_prompt_validation_rules_require_all(self, mock_clusterer):
         )
         assert "Every reference ID must be assigned" in system
 
+    def test_build_prompt_allows_literal_braces(self, mock_clusterer):
+        """Prompt templates with literal braces should not raise formatting errors."""
+        inputs = [(1, "Item {A}"), (2, "Item B")]
+        response_schema = mock_clusterer._build_cluster_schema(SimpleCluster, n_items=2)
+        system, user = mock_clusterer._build_prompt(
+            inputs,
+            response_schema,
+            n_clusters=2,
+            allow_overlap=False,
+            require_all=True,
+            system_prompt=(
+                "Schema: {format}\n"
+                "Clustering: {n_clusters_instruction}\n"
+                "Rules: {validation_rules} | literal {rules}"
+            ),
+            user_prompt="Items:\n{items}\nliteral {notes}",
+        )
+        assert "literal {rules}" in system
+        assert "literal {notes}" in user
+        assert "Item {A}" in user
+
 
 # ============================================================================
 # Validation Tests