ROB-1933: make sure approved or denied tool call triggers a SSE event #1054
Conversation
WalkthroughProcess tool-approval decisions now produces both updated messages and STREAM Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant C as Client
participant S as Server (/chat)
participant L as ToolCallingLLM
participant T as Tool(s)
C->>S: POST /chat (messages, tool_decisions, stream=true)
S->>L: call_stream(messages, tool_decisions)
rect #EFEFEF
note right of L: process_tool_decisions -> (messages, events)
L-->>S: yield TOOL_RESULT events (denied/approved)
end
alt approved
L->>T: invoke tool (user_approved=True)
T-->>L: ToolCallResult(success/data)
L-->>S: yield TOOL_RESULT (approved + result)
else denied
L-->>S: yield TOOL_RESULT (denied)
end
note over L: continue normal LLM streaming
L-->>S: assistant token/events
S-->>C: SSE stream (TOOL_RESULT + tokens)
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Possibly related PRs
Suggested reviewers
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
holmes/core/tool_calling_llm.py (1)
337-379: Avoid 500s on stale/mismatched decisions; preserve tool result order on insert
- Raising when no pending approvals found will 500 the SSE route. Prefer emitting an ERROR event and returning unchanged messages.
- Inserting multiple tool results at the same index reverses their order. Track per-message insert offset to preserve tool_calls order.
- if not pending_tool_calls: - error_message = f"Received {len(tool_decisions)} tool decisions but no pending approvals found" - logging.error(error_message) - raise Exception(error_message) + if not pending_tool_calls: + error_message = ( + f"Received {len(tool_decisions)} tool decisions but no pending approvals found" + ) + logging.warning(error_message) + events.append( + StreamMessage( + event=StreamEvents.ERROR, + data={"message": error_message}, + ) + ) + return messages, events @@ - for tool_call_with_decision in pending_tool_calls: + # Preserve insertion order per assistant message + insert_offsets: dict[int, int] = {} + for tool_call_with_decision in pending_tool_calls: @@ - messages.insert( - tool_call_with_decision.message_index + 1, tool_call_message - ) + idx = tool_call_with_decision.message_index + offset = insert_offsets.get(idx, 0) + messages.insert(idx + 1 + offset, tool_call_message) + insert_offsets[idx] = offset + 1
🧹 Nitpick comments (3)
tests/test_approval_workflow.py (2)
263-274: Fix Ruff ARG005 and RUF005 in test lambda
- Rename unused arg to
_tool_decisions.- Prefer iterable unpacking over concatenation.
- ai.process_tool_decisions = MagicMock( - side_effect=lambda messages, tool_decisions: ( - messages - + [ + ai.process_tool_decisions = MagicMock( + side_effect=lambda messages, _tool_decisions: ( + [ + *messages, { "tool_call_id": "tool_call_123", "role": "tool", "name": "kubectl_delete", "content": "pod 'dangerous-pod' deleted", } - ], + ], [], # Empty list for StreamMessages ) )
409-420: Fix Ruff ARG005 and RUF005 in rejection test lambda
- Rename unused arg to
_tool_decisions.- Use iterable unpacking.
- ai.process_tool_decisions = MagicMock( - side_effect=lambda messages, tool_decisions: ( - messages - + [ + ai.process_tool_decisions = MagicMock( + side_effect=lambda messages, _tool_decisions: ( + [ + *messages, { "tool_call_id": "tool_call_123", "role": "tool", "name": "kubectl_delete", "content": "Tool execution was denied by the user.", } - ], + ], [], # Empty list for StreamMessages ) )holmes/core/tool_calling_llm.py (1)
293-306: Update docstring to reflect new return typeThe function now returns (messages, events) but docstring says “Updated messages list”. Align it to avoid confusion and mypy hinting issues.
- Returns: - Updated messages list with tool execution results + Returns: + A tuple of: + - Updated messages with tool execution results + - Stream events (TOOL_RESULT) emitted for each decision
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
holmes/core/tool_calling_llm.py(6 hunks)server.py(1 hunks)tests/test_approval_workflow.py(2 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/*.py: Use Ruff for formatting and linting (configured in pyproject.toml) for all Python code
Type hints are required; code should pass mypy (configured in pyproject.toml)
ALWAYS place Python imports at the top of the file, not inside functions or methods
Files:
server.pytests/test_approval_workflow.pyholmes/core/tool_calling_llm.py
tests/**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
Only use pytest markers that are defined in pyproject.toml; never introduce undefined markers/tags
Files:
tests/test_approval_workflow.py
tests/**
📄 CodeRabbit inference engine (CLAUDE.md)
Test files should mirror the source structure under tests/
Files:
tests/test_approval_workflow.py
🧬 Code graph analysis (1)
holmes/core/tool_calling_llm.py (5)
holmes/utils/stream.py (2)
StreamMessage(25-27)StreamEvents(15-22)holmes/core/models.py (4)
ToolCallResult(23-60)as_streaming_tool_result_response(50-60)as_tool_call_message(30-36)ToolApprovalDecision(177-181)holmes/core/tracing.py (1)
DummySpan(101-122)holmes/core/tools.py (2)
StructuredToolResult(79-103)StructuredToolResultStatus(52-76)holmes/core/safeguards.py (1)
prevent_overly_repeated_tool_call(78-120)
🪛 Ruff (0.13.3)
tests/test_approval_workflow.py
263-263: Unused lambda argument: tool_decisions
(ARG005)
264-272: Consider iterable unpacking instead of concatenation
Replace with iterable unpacking
(RUF005)
409-409: Unused lambda argument: tool_decisions
(ARG005)
410-418: Consider iterable unpacking instead of concatenation
Replace with iterable unpacking
(RUF005)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (3)
- GitHub Check: llm_evals
- GitHub Check: Pre-commit checks
- GitHub Check: build
🔇 Additional comments (3)
holmes/core/tool_calling_llm.py (2)
684-691: Good: skip redundancy safeguard when user approvedGuarding prevent_overly_repeated_tool_call behind user_approved avoids blocking explicit approvals.
899-912: Pre-processing and streaming decisions: looks correctProcesses decisions, yields TOOL_RESULT events, and carries updated msgs into the loop. Nice.
server.py (1)
367-368: No action needed: ChatRequest.tool_decisions is defined on a Pydantic BaseModel, so incoming lists of dicts are coerced into ToolApprovalDecision instances.
| tool_calls=previous_tool_calls, | ||
| ) | ||
| tool_response = None | ||
| if not user_approved: |
There was a problem hiding this comment.
why we changed it? if the user approved it but it was already ran once we still would like to run it?
There was a problem hiding this comment.
I changed it because I noticed it. My thinking is that if a user actively approves a command then it should run, regardless of any restriction, including repeated commands.
There was a problem hiding this comment.
Not sure i fully agreed with that. maybe worth discuss it with product and make sure?
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (3)
holmes/core/tool_calling_llm.py (3)
292-292: Use consistent type annotation style.The return type mixes lowercase built-in types (
tuple,list) with uppercase typing module types (List,Dict). For consistency with the rest of the codebase which imports and usesListfrom thetypingmodule (line 5), useTuple[List[Dict[str, Any]], List[StreamMessage]].Apply this diff:
- ) -> tuple[List[Dict[str, Any]], list[StreamMessage]]: + ) -> Tuple[List[Dict[str, Any]], List[StreamMessage]]:And add
Tupleto the imports on line 5:-from typing import Dict, List, Optional, Type, Union, Callable, Any +from typing import Dict, List, Optional, Type, Union, Callable, Any, Tuple
303-303: UseList[StreamMessage]for consistency.The variable declaration uses lowercase
list[StreamMessage]while the codebase consistently usesListfrom thetypingmodule.Apply this diff:
- events: list[StreamMessage] = [] + events: List[StreamMessage] = []
899-910: Mixed union syntax.Line 899 uses the newer union syntax
List[ToolApprovalDecision] | Nonewhile the rest of the codebase usesOptional[List[ToolApprovalDecision]]. While both work in Python 3.10+, consistency with the existing codebase style is preferred.The logic for processing tool decisions and yielding events is correct and aligns with the PR objectives.
Apply this diff for consistency:
- tool_decisions: List[ToolApprovalDecision] | None = None, + tool_decisions: Optional[List[ToolApprovalDecision]] = None,
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
holmes/core/tool_calling_llm.py(6 hunks)
🧰 Additional context used
📓 Path-based instructions (1)
**/*.py
📄 CodeRabbit inference engine (CLAUDE.md)
**/*.py: Use Ruff for formatting and linting (configured in pyproject.toml) for all Python code
Type hints are required; code should pass mypy (configured in pyproject.toml)
ALWAYS place Python imports at the top of the file, not inside functions or methods
Files:
holmes/core/tool_calling_llm.py
🧬 Code graph analysis (1)
holmes/core/tool_calling_llm.py (5)
holmes/utils/stream.py (2)
StreamMessage(25-27)StreamEvents(15-22)holmes/core/models.py (4)
ToolCallResult(23-60)as_streaming_tool_result_response(50-60)as_tool_call_message(30-36)ToolApprovalDecision(177-181)holmes/core/tracing.py (1)
DummySpan(101-122)holmes/core/tools.py (2)
StructuredToolResult(79-103)StructuredToolResultStatus(52-76)holmes/core/safeguards.py (1)
prevent_overly_repeated_tool_call(78-120)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (3)
- GitHub Check: Pre-commit checks
- GitHub Check: llm_evals
- GitHub Check: build
🔇 Additional comments (2)
holmes/core/tool_calling_llm.py (2)
342-378: LGTM! Tool approval logic correctly emits SSE events.The implementation properly handles both approved and denied tools, generating
TOOL_RESULTevents in both cases and deriving the tool call message from the result. This aligns perfectly with the PR objective to ensure SSE messages are sent for tool results.
684-690: LGTM! Approved tools correctly bypass the repeated-call safeguard.When
user_approved=True, the safeguard is skipped, which makes sense—if a user explicitly approves a tool, it should run regardless of previous executions.
2 participants
Ensures SSE messages are sent to the UI for tool results when tools are approved or denied. Otherwise the UI does not get updated.