Skip to content

fix(planning): forbid questions embedded in interview recommendations and warn on add #7760

Closed

Description

@kyle-sexton

Parent

Refs #5893

Depends on

  • none: on the frontier now

Outcome

A recommendation never hides a second question. Reported case: Q123's recommendation ended "Also tell me which rung you intend"; the user pressed Accept with no note, which settled the headline but not the embedded question, so the session had to add another question. The skill forbids an embedded question (ask it as its own question), and add warns on one without refusing.

Owner decision recorded on #5893 (2026-10-12): "Item 5: how to handle a question embedded in a recommendation?" → Forbid it in the skill text and have add warn.

Key interfaces

  • Interview skill question-authoring rules (surface.md or SKILL.md, wherever recommendations are specified).
  • round.py add (and revise if it sets the recommendation). The detection heuristic is chosen in this item; it warns and never refuses.

Done when

  • The skill text states that a recommendation carries no question to the user.
  • add warns on a recommendation that asks the user something, such as the reported "Also tell me which rung you intend" (test).
  • add does not warn on an ordinary recommendation (test).
  • The planning Python and surface test suites pass.

Out of scope

  • Refusing on a text heuristic.

Activity

  1. added theissue type on Oct 17, 2026
  2. added
    priority: mediumReal value, no hard deadline; normal backlog flow.
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Oct 17, 2026
  3. kyle-sexton commented on Oct 17, 2026

    @kyle-sexton
    MemberAuthor

    Ratified in the owner's batch of decomposed children (2026-10-17).

  4. removed
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Oct 17, 2026
  5. kyle-sexton commented on Oct 18, 2026

    @kyle-sexton
    MemberAuthor

    Agent Brief

    Type: Bug
    Summary: Forbid questions embedded in interview recommendations, and warn on one at add.

    Current behavior:
    A recommendation can end with a second question ('Also tell me which rung you intend'); Accept settles the headline but not the embedded question.

    Desired behavior:
    The interview surface rules state that a recommendation carries no question to the user; a second question is asked as its own question. add (and revise when it sets the recommendation) warns, never refuses, on a recommendation that asks the user something. The detection heuristic is chosen in this change.

    Key interfaces:

    • Interview surface question-authoring rules for recommendation (the context surface spoke, not SKILL.md).
    • round.py add and revise: a warn-only check.

    Acceptance criteria:

    • The surface spoke states a recommendation carries no question to the user (read from the diff); the interview SKILL.md is not edited.
    • test_round.py gains a case where add warns on 'Also tell me which rung you intend' and a case where an ordinary recommendation draws no warning, and scripts/run-test-file.sh plugins/planning/surface/test_round.py exits 0.
    • A planning changelog fragment is added and scripts/check-changelog-fragments.sh exits 0.

    Out of scope:

    • Refusing on a text heuristic.
    • The interview SKILL.md.

    This was generated by AI during backlog shaping. Held, not admitted: this brief is drafted and checked, but the issue lane would stop on it today. It is admitted, by adding agent-ready, once the lane can run the skill-quality check scripts (Automation Lanes' PR #7912). Held under the owner's 2026-10-18 lane-narrowing course correction.

  6. added
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    work-class: scopedA briefed fix or small feature; blast radius bounded by the brief, tests exist.
    and removed
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Oct 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.priority: mediumReal value, no hard deadline; normal backlog flow.work-class: scopedA briefed fix or small feature; blast radius bounded by the brief, tests exist.

    Type

    Fields

    No fields configured for Bug.

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions