Skip to content

Evaluate an early ambiguity preflight for autonomous prompt execution #267

Description

Problem

The instruction-fidelity guardrail handles material ambiguity when it is encountered, and bootstrap.md proactively scans substantive parameter values during collection. In autonomous sessions, however, an ambiguity that materially affects downstream decisions may only be noticed after execution has started, when user input is unavailable.

Proposed direction

Evaluate adding an explicit ambiguity preflight near the beginning of prompt execution. The preflight should:

  • identify material ambiguities that could affect later decisions;
  • ask targeted questions when interaction is permitted;
  • when interaction is unavailable, preserve viable alternatives as [OPEN QUESTION] rather than silently choosing one;
  • avoid blocking on minor or non-material uncertainty;
  • compose cleanly with existing template phases and approval gates.

Context

Follow-up from PR #266 review feedback by Andrew Beltrano (@abeltrano). This is intentionally separate from the universal instruction-fidelity guardrail introduced by that PR so the timing and autonomous-execution semantics can be designed without expanding the current change.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions