Skip to content

Should each verification check record the files it covers, so upkeep can classify a change? #15656

Closed

Description

@kyle-sexton

Question

Should each verification check record the files it covers and a sync commit, so upkeep can classify a change as no effect, drift, intended change or possible break?

From the talk

Chapter 7, slides 25-27: runtime and formal verification artifacts; her pstack plugin keeps checks tied to the code they describe.

Today and the gap

A check can keep passing while it describes code that no longer exists. /app-control:map refresh brings a control skill back in line, but no record says which files each check covers. Basis: pstack's coverage-record idea, recorded in docs/upstream/cursor-pstack.md.

Recommendation

Proof of concept (menu priority P1), extending /app-control:map refresh. Basis: the gap above.

Alternatives

Rely on refresh re-deriving everything each run; record coverage only for formal models.

Done would add

A coverage field in the control-skill contract and a classification step in refresh. Dogfood: this repository's own control skill (.claude/skills/control-claude-code-plugins/) gains the record and a refresh run classifies a real change.

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: Medium
    Complexity: Medium
    Uncertainty: Knowable
    Flags: none
    Blocked by: none

    Understood: you propose that each verification check record the files it covers plus a sync commit. /app-control:map refresh could then classify a change as no effect, drift, intended change or possible break. The issue picks a recommendation (a proof of concept extending refresh) and names its alternatives. Done means a coverage field in the control-skill contract, a classification step in refresh, and a dogfood run on this repository's own control skill.

    Nothing essential is missing. The classification rules, and where the coverage record is stored, are implementation choices to settle during the work.

    Possibly related, though not a duplicate: #12 on other pstack ideas. No merged pull request appears to cover this.

    Automated triage. A maintainer reviews it.

  4. melodic-automation-lanes commented on Nov 12, 2026

    @melodic-automation-lanes
    Contributor

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

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

    Reason: intent

  5. kyle-sexton commented on Nov 13, 2026

    @kyle-sexton
    MemberAuthor

    Resolved: already covered for runtime checks; for formal models it becomes a requirement on the proof of concept. No new build item and no new field in the control-skill contract.

    From the talk. Chapter 7, slides 25-27 (runtime and formal verification artifacts). The coverage record and sync commit themselves are not in the talk; they come from her plugin's formal-verification skills (pstack/skills/create-formal-verification-skill/SKILL.md:33, pstack/skills/maintain-formal-verification-skill/SKILL.md:28-36 in cursor/plugins), not her runtime ones.

    Runtime checks: already have.

    • Each feature in a control skill records the files it covers as sources globs, and /app-control:drive affected matches a change against them (plugins/verification/skills/confirm/SKILL.md:166).
    • /app-control:map refresh keeps a record of its last pass per control skill (plugins/app-control/skills/map/SKILL.md:194, :212-213) and triages each difference as doc drift, driver gap or product regression (:209).
    • So this item's premise, that no record says which files each check covers, does not hold for runtime checks. A second coverage field beside sources would duplicate it.

    Formal models: carried to the proof of concept. Model the loop-lane claim protocol in TLA+ as one bounded proof of concept now requires a coverage record and a sync commit for any model worth keeping, with her four change classes. Whether formal verification becomes its own skill is decided in that item's interview, by responsibility, not presumed.

    Owner, 2026-11-12: "Yeah, go with recommendations."

    Not verified: what the refresh skip record keys on (its script was not read).

  6. 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.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