Nu Hoang
@htn274 · Ph.D. student @ A2I2, Deakin University. Working on Causal Reasoning in AI.
diplomatic and thorough
Thorough and methodical reviewer who seeks deep understanding before making suggestions. Approaches reviews with curiosity and humility, often asking clarifying questions and admitting when they need more context or are new to the codebase.
20
Comments
9
PRs
3
Repos
602
Avg Chars
3
Harshness
Personality
Humble and self-aware
Detail-oriented and methodical
Collaborative and respectful
Patient with complex problems
Clear communicator
Solution-focused
Curious and inquisitive
Professional and courteous
Greatest Hits
"This is a good catch. But I don't think your fix is correct."
"Based on my understanding, [question]. Am I correct?"
"As I am new to onboarding, any additional clarification for this ticket would be greatly appreciated."
"just ignore it at the moment. @coderabbitai ignore"
"My solution: We should check BaseModel here and throw an error"
Focus Areas
- code correctness
- data validation
- error handling
- architectural understanding
- proper model usage
- session management
- prompt engineering
Common Phrases
"Based on my understanding"
"I think this"
"would be"
"should be"
"Am I correct?"
"Thanks for"
"Hi [Name]"
"This is a good catch"
"I don't think your fix is correct"
"just ignore"
"Please see"
"for details"
"My solution:"
"As I am new to"
Sentiment Breakdown
Review Outcomes
Most Reviewed Authors
AI Persona Prompt
You are @htn274, a thoughtful and methodical code reviewer who approaches each review with genuine curiosity and humility. You're relatively new to some codebases but aren't afraid to ask clarifying questions. Start comments with 'Hi [Name],' and end with courteous sign-offs like 'Thanks!' or 'Best,'. When you spot issues, acknowledge good catches with 'This is a good catch' before explaining why the solution might not be correct. You frequently say 'Based on my understanding' and ask 'Am I correct?' to confirm your interpretation. You focus heavily on data validation, proper model usage (especially Pydantic BaseModel), and architectural correctness. When suggesting solutions, prefix with 'My solution:' and provide specific code examples. You're patient with complex problems and will defer issues to 'next iteration' when needed. Use phrases like 'just ignore' for irrelevant code and '@coderabbitai ignore' for automated tools. You provide detailed explanations with links to specific code lines and attach log files when demonstrating issues. Your tone is professional but warm, and you genuinely want to understand the codebase better while helping maintain code quality.
Recent Comments (13 total)
Hi @gauravreddy08,
This is a good catch.
But I don't think your fix is correct.
This is because we expect the data MUST be a Pydantic model.
**My solution:** We should check BaseModel here and throw an error if the generated code doesn’t return a Base model.
https://github.com/tinyfish-io/web-agent/blob/c95a34b28b805f5d3a3c74a9826753dcd8a40201/web-agent-core/src/web_agent_core/steps/ex
Based on my understanding, the error raising in this step will be sent back to the linear model. Am I correct?
I think just add these lines before returning `data`:
```python
if not isinstance(data, BaseModel):
raise ValueError("The 'extract_data' function did not return a Pydantic BaseModel instance.")
```
We already handled this logic in `llm_codifier.py`.
https://github.com/tinyfish-io/web-agent/blob/222861d4ef6064b253710a48ff760d4ec309c6be/web-agent-core/src/web_agent_core/steps/extract_data/llm_codifier.py#L85-L91
Should be: `step = state.execution.step`
Same as non-codified: `state.execution.step.codified`
I think this line is not correct about extracting data. The goal of the extract data step is to get the data in a structured way. This component does not include formatting data (which I think belongs to transform data).
As a result, the planner makes this type of plan with the prompt: "Navigate to the 5th story on https://news.ycombinator.com and open it, return the title and brief summary. In J
Same as above. Need to revise the prompt for a better description of "extract data step".
This code is a part of an alternative approach, which is currently not used.
Therefore, just ignore it at the moment. @coderabbitai ignore
@coderabbitai ignore
i! I've just checked it and got the error here.
Please see the log below for details.
```
Loaded step from: /Users/nunu/EWA-eco/web-agent/web-agent-core/codified/steps_fedex.json
Title: Navigate to Fedex and find shipping price for a 3lb package
Codified: True
Children: 16
1. NavigateStep: Navigate to FedEx homepage (codified=True)
2. ClickStep: Accept cookies (codified=True)
3.
Hi Pasha,
Thanks for your prompt feedback.
I would like to reiterate my understanding of this ticket: We would like to eliminate the concept of an offline session ID entirely.
Upon reviewing the code, I observe that the offline session ID is utilized for replay sessions. Does this imply that we should also remove the session ID when initializing a session in this line of code?
https://g
web-agent-core/src/web_agent_core/steps/extract_data/step.py[view]