Skip to content

Should we adopt her two one-line verification rules, and in which skills? #15657

Closed

Description

@kyle-sexton

Question

Should we adopt her two one-line verification rules: "replay every model-found bug against the real code; if it does not occur, assume the model is wrong", and "suspect the observation method before suspecting the system"?

From the talk

Chapter 7, slides 27 and 32: formal verification; both lines come from her pstack verification skills.

Today and the gap

Neither is recorded as taken or declined in docs/upstream/cursor-pstack.md.

Recommendation

Take both (immediate takeaway, menu priority P1): the first in /prototype:pressure-test or /verification:confirm, the second in /debugging:debug. The formal-verification work item already applies the first: Model the loop-lane claim protocol in TLA+ as one bounded proof of concept. Basis: judgment; both are cheap and match existing skill intent.

Alternatives

Decline with a reason in the upstream record; take only one.

Done would add

One line in each named skill, with an eval case showing it changes behavior, and a row in the upstream record. Dogfood: the eval runs in this repository.

Activity

  1. added
    wayfind: interviewWayfind decision item: contract/requirements call the user must make; human in the loop.
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Nov 12, 2026
  2. added theissue type on Nov 12, 2026
  3. github-actions commented on Nov 12, 2026

    @github-actions
    Contributor

    Size: Small
    Complexity: Low
    Uncertainty: Known
    Flags: intent
    Blocked by: none

    Understood: you ask whether to add two one-line verification rules to named skills, with an eval case each and a row in the upstream record. The issue leaves the decision, and the first rule's target skill, open. It lists decline or take-one as alternatives. No duplicate or merged fix found; #15648 is related and already applies the first rule.

    1. Adopt both rules? RECOMMENDED: yes, take both. Basis: the issue's own judgment that they are cheap and match existing skill intent.
    2. Which skill gets rule one: /prototype:pressure-test or /verification:confirm? RECOMMENDED: /verification:confirm, since replaying a finding against real code is confirmation. Basis: judgment; the issue names both and picks neither.

    To answer, post a comment that starts with this line:

    **Decision recorded.** <answer>
    

    Automated triage. A maintainer reviews it.

  4. added
    needs-owner-decisionAwaiting a human or maintainer judgment call.
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Nov 12, 2026
  5. melodic-automation-lanes commented on Nov 12, 2026

    @melodic-automation-lanes
    Contributor

    The triage verdict left this issue at needs-owner-decision:

    • flagged: intent
    • held by needs-owner-decision
    • held by needs-human
    • no lanes App verdict comment shows whether an earlier verdict was flagged

    Reason: intent

  6. kyle-sexton commented on Nov 13, 2026

    @kyle-sexton
    MemberAuthor

    Resolved: take both rules, each in the one place that owns it.

    From the talk. Chapter 7, slides 27 and 32 (formal verification and statecharts; transcript 1:12:08 to 1:12:39 and 1:14:05). The two lines themselves are from her plugin, not the talk.

    Rule Her source (cursor/plugins, MIT) Where it goes
    Replay every bug a model finds against the real code; "If a bug does not occur, think that the model is wrong until you show that it is not" pstack/skills/statechart/SKILL.md:82; the same replay is step 5 of maintain-formal-verification-skill The formal-verification plugin: the chart skill's hand-off (slice 1) and the refresh action (slice 3). It applies only to models.
    "When verification fails, suspect the observation method before suspecting the system" pstack/skills/principle-prove-it-works/SKILL.md:16 /verification:confirm, once, with one eval case: Suspect the observation method first when a verification fails in verification:confirm

    Why confirm and not debug. The second rule is about reading a failed verification, and confirm is the skill that judges one. Stating it in /debugging:debug as well would put one rule in two places.

    Owner, 2026-11-12: "Yes, let's go with recommendations there."

    Basis: a search of this marketplace's plugins for the second rule found no equivalent (run by a comparison agent on 2026-11-12, not repeated by hand); the first rule's lines were read by that agent and relayed.

  7. kyle-sexton commented on Nov 13, 2026

    @kyle-sexton
    MemberAuthor

    Resolved; see the resolution comment.

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

    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.needs-owner-decisionAwaiting a human or maintainer judgment call.needs-triageNot yet classified. Floor until a type and one priority tier are set.wayfind: interviewWayfind decision item: contract/requirements call the user must make; human in the loop.

    Type

    Fields

    Work status

    needs-human

    Priority

    medium

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions