Unable to create custom callback for pandas agent during intermediate execution

[dineshresearch]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).

Example Code

None

Error Message and Stack Trace (if applicable)

None

Description

• I am trying to implement a guardrail for pandas agent in such a way that
• I want to monitor the pandas REACT agent after passing the input, prompt and during thought phase the agent will generate code.
• Using callbacks, I want to analyze the generated code and stop the code execution if it is malicious without proceeding to further steps.

System Info

0.3.2

[eyurtsev]

Callbacks are generally meant to run in the background, so the strategy itself doesn't make a lot of sense given the design patterns for callbacks. However, you can use run_inline = True on the callback handler to force it to run_inline (so it won't be in the background).

If you need to get this working in a production environment, you should change the overall strategy.

• either sandbox the agent completely and assume that it can be used maliciously but given that it's sandboxed you may not care
• load the content in duckdb and have the agent issue SQL directly and configure duckdb safely.
• write a small dsl that issues a list of allowed commands against a dataframe