Skip to content

Agent should stop/discourage destructive behavior #22672

@abhipatel12

Description

@abhipatel12

In certain cases (complex git operations, branch management, etc), the model can occasionally use commands like git reset or --force when a safer alternative is possible.

Additionally, when maintaining resources such as DBs, etc, we should ensure that the model understands the dangers of modifying resources without a way of reverting any accidental destructive behavior.

Ideally, we can capture this type of behavior and add behavioral evals to ensure it does not happen.

Metadata

Metadata

Assignees

No one assigned

    Labels

    area/agentIssues related to Core Agent, Tools, Memory, Sub-Agents, Hooks, Agent Qualitykind/customer-issueIssues that were reported by customerspriority/p2Important but can be addressed in a future release.status/bot-triagedworkstream-rollupLabel used to tag epics and features that are associated with one of the three primary workstreams🔒 maintainer only⛔ Do not contribute. Internal roadmap item.

    Type

    No fields configured for Bug.

    Projects

    Status

    No status

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions