Skip to content

Debug mode for tactics #459

Description

@Kaptch

During the Iris workshop, it was brought up that if someone wants to apply LLMs to help with Iris proofs, it would be beneficial to have a verbose output mode. I haven't thought about the design, so we can discuss it here. Perhaps, one could add a flag that in case of errors makes tactics dump the Iris goal in a more structured way. Plus, it would be nice if the output can be used to state intermediate goals (that is, output contains a valid Entails statement).

If anyone has more ideas on what would be helpful, feel free to propose other ideas. I haven't used LLMs for any big scale things, so I am not sure what would be helpful or not.

Possibly, there can be also some structured summary of Iris tactics?

Metadata

Metadata

Assignees

No one assigned

    Labels

    ImprovementNot a bug, but something can still be improvedexperimentIdeas for features that may or may not workquestionFurther information is requested

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions