Add cost cap at 80% of revenue; LlmResult dataclass; rename get_credit_price

constants/agent.py

@@ -1,2 +1,4 @@
+# Stop agent loop when LLM cost reaches this fraction of revenue (credit_cost_usd)
+COST_CAP_RATIO = 0.8
 MAX_ITERATIONS = 30
 MAX_PLANNING_ITERATIONS = 20

scripts/git/pre_commit_hook.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# Git pre-commit hook. No stashing - runs on the working directory as-is.
+# Git pre-commit hook. Does NOT auto-stage — user handles staging.
 # Install: ln -sf ../../scripts/git/pre_commit_hook.sh .git/hooks/pre-commit
 set -uo pipefail
 
@@ -35,17 +35,12 @@ python3 schemas/supabase/generate_types.py && git add schemas/supabase/
 # Get staged Python files (excluding deleted, .venv, schemas)
 STAGED_PY_FILES=$(git diff --cached --name-only --diff-filter=d -- '*.py' | grep -v '^\.\?venv/' | grep -v '^schemas/')
 
-# Format and auto-fix staged Python files
+# Format staged Python files (user re-stages if needed)
 if [ -n "$STAGED_PY_FILES" ]; then
     # shellcheck disable=SC2086
     black $STAGED_PY_FILES
-    # shellcheck disable=SC2086
-    git add $STAGED_PY_FILES
-
     # shellcheck disable=SC2086
     ruff check --fix $STAGED_PY_FILES
-    # shellcheck disable=SC2086
-    git add $STAGED_PY_FILES
 fi
 
 # Markdownlint for staged .md files

services/chat_with_agent.py

@@ -32,10 +32,11 @@
 
 
 @dataclass
-class AgentResult:
+class AgentResult:  # pylint: disable=too-many-instance-attributes
     messages: list[MessageParam]
     token_input: int
     token_output: int
+    cost_usd: float
     is_completed: bool
     completion_reason: str
     p: int
@@ -67,12 +68,7 @@ async def chat_with_agent(
         logger.info("Using model: %s", current_model)
 
         try:
-            (
-                response_message,
-                tool_calls,
-                token_input,
-                token_output,
-            ) = chat_with_model(
+            llm_result = chat_with_model(
                 messages=messages,
                 system_content=system_message,
                 tools=tools,
@@ -125,24 +121,25 @@ async def chat_with_agent(
             )
 
     # Return if no tool calls (agent returned text without calling a tool)
-    if not tool_calls:
-        logger.info("No tools were called. Response: %s", response_message)
-        messages.append(response_message)
+    if not llm_result.tool_calls:
+        logger.info("No tools were called. Response: %s", llm_result.assistant_message)
+        messages.append(llm_result.assistant_message)
         return AgentResult(
             messages=messages,
-            token_input=token_input,
-            token_output=token_output,
+            token_input=llm_result.token_input,
+            token_output=llm_result.token_output,
+            cost_usd=llm_result.cost_usd,
             is_completed=False,
             completion_reason="",
             p=p,
             is_planned=False,
         )
 
     # Append assistant message before processing tool calls
-    messages.append(response_message)
+    messages.append(llm_result.assistant_message)
 
     # Extract text from the assistant message for completion context
-    content = response_message["content"]
+    content = llm_result.assistant_message["content"]
     if isinstance(content, str):
         assistant_text = content
     else:
@@ -156,11 +153,11 @@ async def chat_with_agent(
     tool_result_blocks: list[ToolResultBlockParam] = []
     log_msgs: list[str] = []
     is_completed = False
-    num_tool_calls = len(tool_calls)
+    num_tool_calls = len(llm_result.tool_calls)
     logger.info("Processing %d tool call(s)", num_tool_calls)
 
     # pylint: disable-next=too-many-nested-blocks
-    for i, tc in enumerate(tool_calls, start=1):
+    for i, tc in enumerate(llm_result.tool_calls, start=1):
         tool_use_id = tc.id
         tool_name = tc.name
         tool_args = tc.args
@@ -505,7 +502,7 @@ async def chat_with_agent(
     if log_msgs:
         update_comment(
             body=create_progress_bar(
-                p=p + 5 * len(tool_calls), msg="\n".join(log_messages)
+                p=p + 5 * len(llm_result.tool_calls), msg="\n".join(log_messages)
             ),
             base_args=base_args,
         )
@@ -520,10 +517,11 @@ async def chat_with_agent(
 
     return AgentResult(
         messages=messages,
-        token_input=token_input,
-        token_output=token_output,
+        token_input=llm_result.token_input,
+        token_output=llm_result.token_output,
+        cost_usd=llm_result.cost_usd,
         is_completed=is_completed,
         completion_reason=assistant_text,
-        p=p + 5 * len(tool_calls),
+        p=p + 5 * len(llm_result.tool_calls),
         is_planned=False,
     )

services/claude/chat_with_claude.py

@@ -1,36 +1,26 @@
-# Standard imports
-from dataclasses import dataclass
 import time
 
-# Third party imports
 from anthropic import AuthenticationError
 from anthropic._exceptions import OverloadedError
 from anthropic.types import MessageParam, ToolUnionParam, ToolUseBlock
 
-# Local imports
 from constants.claude import CONTEXT_WINDOW, MAX_OUTPUT_TOKENS
 from constants.models import ClaudeModelId
 from services.claude.client import claude
-from services.claude.strip_strict_from_tools import strip_strict_from_tools
 from services.claude.exceptions import (
     ClaudeAuthenticationError,
     ClaudeOverloadedError,
 )
 from services.claude.remove_outdated_file_edit_attempts import (
     remove_outdated_file_edit_attempts,
 )
+from services.claude.strip_strict_from_tools import strip_strict_from_tools
 from services.claude.trim_messages import trim_messages_to_token_limit
+from services.llm_result import LlmResult, ToolCall
 from services.supabase.llm_requests.insert_llm_request import insert_llm_request
 from utils.error.handle_exceptions import handle_exceptions
 
 
-@dataclass
-class ToolCall:
-    id: str
-    name: str
-    args: dict | None
-
-
 @handle_exceptions(raise_on_error=True)
 def chat_with_claude(
     messages: list[MessageParam],
@@ -126,7 +116,7 @@ def chat_with_claude(
     # Combine system message with user messages for logging
     system_msg: MessageParam = {"role": "user", "content": system_content}
     full_messages = [system_msg, *messages]
-    insert_llm_request(
+    llm_record = insert_llm_request(
         usage_id=usage_id,
         provider="claude",
         model_id=model_id,
@@ -137,10 +127,12 @@ def chat_with_claude(
         response_time_ms=response_time_ms,
         created_by=created_by,
     )
-
-    return (
-        assistant_message,
-        tool_calls,
-        token_input,
-        token_output,
+    cost_usd = llm_record["total_cost_usd"] if llm_record else 0.0
+
+    return LlmResult(
+        assistant_message=assistant_message,
+        tool_calls=tool_calls,
+        token_input=token_input,
+        token_output=token_output,
+        cost_usd=cost_usd,
     )

services/claude/test_chat_with_claude.py

@@ -13,6 +13,7 @@ def test_chat_with_claude_success(mock_claude, mock_insert_llm_request):
     mock_response = Mock()
     mock_response.content = [Mock(type="text", text="Hello! How can I help you?")]
     mock_response.usage = Mock(output_tokens=15)
+    mock_insert_llm_request.return_value = {"total_cost_usd": 0.05}
 
     mock_claude.messages.create.return_value = mock_response
     mock_claude.messages.count_tokens.return_value = Mock(input_tokens=20)
@@ -31,13 +32,13 @@ def test_chat_with_claude_success(mock_claude, mock_insert_llm_request):
         created_by="4:test-user",
     )
 
-    assistant_message, tool_calls, token_input, token_output = result
-    assert assistant_message["role"] == "assistant"
-    content = cast(list, assistant_message["content"])
+    assert result.assistant_message["role"] == "assistant"
+    content = cast(list, result.assistant_message["content"])
     assert content[0]["text"] == "Hello! How can I help you?"
-    assert not tool_calls
-    assert token_input == 20
-    assert token_output == 15
+    assert not result.tool_calls
+    assert result.token_input == 20
+    assert result.token_output == 15
+    assert result.cost_usd == 0.05
 
     mock_insert_llm_request.assert_called_once()
     call_args = mock_insert_llm_request.call_args[1]
@@ -88,11 +89,10 @@ def test_chat_with_claude_with_tool_use(mock_claude, mock_insert_llm_request):
         created_by="4:test-user",
     )
 
-    _, tool_calls, _, _ = result
-    assert len(tool_calls) == 1
-    assert tool_calls[0].id == "tool_123"
-    assert tool_calls[0].name == "test_function"
-    assert tool_calls[0].args == {"param": "value"}
+    assert len(result.tool_calls) == 1
+    assert result.tool_calls[0].id == "tool_123"
+    assert result.tool_calls[0].name == "test_function"
+    assert result.tool_calls[0].args == {"param": "value"}
 
     mock_insert_llm_request.assert_called_once()
 
@@ -116,8 +116,7 @@ def test_chat_with_claude_no_usage_response(mock_claude, mock_insert_llm_request
         created_by="4:test-user",
     )
 
-    _, _, _, token_output = result
-    assert token_output == 0  # output tokens should be 0 when no usage info
+    assert result.token_output == 0  # output tokens should be 0 when no usage info
     mock_insert_llm_request.assert_called_once()
 
 

services/google_ai/chat_with_google.py

@@ -1,7 +1,6 @@
 # Standard imports
 import time
 import uuid
-from dataclasses import dataclass
 
 # Third-party imports
 from anthropic.types import MessageParam, ToolUnionParam
@@ -12,18 +11,12 @@
 from services.google_ai.client import get_google_ai_client
 from services.google_ai.convert_messages import convert_messages_to_google
 from services.google_ai.convert_tools import convert_tools_to_google
+from services.llm_result import LlmResult, ToolCall
 from services.supabase.llm_requests.insert_llm_request import insert_llm_request
 from utils.error.handle_exceptions import handle_exceptions
 from utils.logging.logging_config import logger
 
 
-@dataclass
-class ToolCall:
-    id: str
-    name: str
-    args: dict | None
-
-
 @handle_exceptions(raise_on_error=True)
 def chat_with_google(
     messages: list[MessageParam],
@@ -105,7 +98,7 @@ def chat_with_google(
     # Log to Supabase
     system_msg: MessageParam = {"role": "user", "content": system_content}
     full_messages = [system_msg, *messages]
-    insert_llm_request(
+    llm_record = insert_llm_request(
         usage_id=usage_id,
         provider="google",
         model_id=model_id,
@@ -116,6 +109,7 @@ def chat_with_google(
         response_time_ms=response_time_ms,
         created_by=created_by,
     )
+    cost_usd = llm_record["total_cost_usd"] if llm_record else 0.0
 
     logger.info(
         "Google AI response: model=%s, input_tokens=%d, output_tokens=%d, tool_calls=%d",
@@ -125,9 +119,10 @@ def chat_with_google(
         len(tool_calls),
     )
 
-    return (
-        assistant_message,
-        tool_calls,
-        token_input,
-        token_output,
+    return LlmResult(
+        assistant_message=assistant_message,
+        tool_calls=tool_calls,
+        token_input=token_input,
+        token_output=token_output,
+        cost_usd=cost_usd,
     )