Skip to content

claude-ops: record the missing #4050 acceptance evidence (unattended cloud transcript and self-check exit) #5302

Description

@kyle-sexton

Filed from the 2026-09-29 audit of the Cursor agent run. Follow-up to #4050. An agent lane must not pick this up: criterion 3 needs a real cloud session, and a transcript written by an agent would be fabricated.

What #4050 delivered

The suggest sentences in three skills and the baked store rows.

What is unmet

  • Criterion 3. No cloud-session transcript of /claude-ops:audit-install-state unattended exists that shows the final section recording the /doctor suggestion without asking.
  • Criterion 2 as written. overlap.py self-check --upstream-sha ed404106fcd80ba98ecb7c851e531dcb626d13b7 is expected to exit 0 but exits 3, because of two advisories: the recorded extraction versions 2.1.232 to 2.1.269 differ from build 2.1.284, and the recorded upstream commit d7dbd9a differs from ed404106. The owner decides whether exit 3 (degraded, a passing run per the audit-native-overlap SKILL.md) is accepted, or whether the store observations are refreshed (core-docs fix group).

Measured 2026-09-29 on the fix branch merged with origin/main: check-skill.sh plugins/claude-ops/skills gave 13 passed, 0 failed. In plugins/claude-ops/skills/audit-install-state/SKILL.md, grep -c "If /doctor is available in your session (" gave 1 and grep -c unattended gave 3.

Operator steps

  1. In a Claude Code cloud session with the claude-ops plugin enabled, run /claude-ops:audit-install-state unattended.
  2. Copy the report's final section and post it on this issue.
  3. Expected: it names the /doctor suggestion and never asks a question.
  4. Decide the exit-3 question above and comment the decision here.

Acceptance criteria

  • The transcript's final section is posted here and names the /doctor suggestion without asking.
  • The owner recorded whether self-check exit 3 is accepted or the store observations get refreshed.

Activity

  1. added
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Sep 29, 2026
  2. added
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Sep 29, 2026
  3. kyle-sexton commented on Sep 29, 2026

    @kyle-sexton
    ContributorAuthor

    Owner decision (2026-09-29): Run the cloud session and accept exit 3. The owner will run /claude-ops:audit-install-state unattended in a real cloud session and post its final section here. Self-check exit 3 (degraded, a passing run per audit-native-overlap SKILL.md) is accepted, so the store observations are not refreshed.

    Next: the issue stays open with needs-human until the owner posts the cloud transcript. Once posted, an agent can tick the acceptance boxes and close the #4050 follow-up.

  4. kyle-sexton commented on Oct 1, 2026

    @kyle-sexton
    ContributorAuthor

    Owner decision (2026-10-01): approved the recommendation ("go with recommended", the owner's reply to the batch decision page that listed this question and its recommendation).

    Question: how to close the missing unattended cloud transcript for #5302 (run it, wait until a date, or waive it).

    Chosen option A: keep the 2026-09-29 decision. The owner runs /claude-ops:audit-install-state unattended in a cloud session with claude-ops enabled and posts the report's final section here.

    Next: the issue stays open with needs-human until that transcript is posted. Then an agent checks that it names the /doctor suggestion and asks nothing, ticks both acceptance boxes (the second citing the 2026-09-29 exit-3 decision), removes needs-human, and closes the issue.

  5. kyle-sexton commented on Oct 4, 2026

    @kyle-sexton
    ContributorAuthor

    This was generated by AI during triage.

    Triage: this stays with needs-human. The missing evidence needs a real cloud session that the owner runs; an agent cannot produce it.

    One correction to the operator steps: the claude-ops plugin was renamed to harness-ops on 2026-10-01 (#5760), so /claude-ops:audit-install-state no longer exists. The renamed skill keeps the same unattended contract: under unattended it records the /doctor suggestion in the report's final section and does not ask.

    Agent Brief

    Type: Task
    Summary: Record the cloud-session evidence #4050's criterion 3 still lacks, then close this follow-up.

    Current behavior:
    No transcript exists of the audit-install-state skill running unattended in a Claude Code cloud session. Criterion 2 is settled: the owner accepted self-check exit 3 (degraded, a passing run) on 2026-09-29 and confirmed it on 2026-10-01.

    Desired behavior:

    1. The owner, in a Claude Code cloud session with the harness-ops plugin enabled, runs /harness-ops:audit-install-state unattended.
    2. The owner posts the report's final section here.
    3. An agent then checks that the posted section names the /doctor suggestion and that the run asked no question, ticks both acceptance boxes (the second citing the 2026-09-29 exit-3 decision), removes needs-human, and closes this issue.

    Why a human must act: capability blocker. A real cloud session must be run by the owner; a transcript written by an agent would be fabricated.

    Acceptance criteria:

    • The final section of an unattended cloud-session run of the harness-ops audit-install-state skill is posted here, names the /doctor suggestion, and the run asked no question.
    • The second acceptance box cites the recorded exit-3 decision.

    Out of scope:

    • Refreshing the native-surface store observations (declined in the exit-3 decision).
    • Any change to the audit-install-state skill.

    Basis: verified, the harness-ops audit-install-state skill's unattended handling on main; the #5760 rename commit; owner comments of 2026-09-29 and 2026-10-01.

  6. added
    priority: lowNice-to-have, cosmetic, or speculative; opportunistic.
    and removed
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Oct 4, 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

    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.priority: lowNice-to-have, cosmetic, or speculative; opportunistic.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions