playbooks: repo-sweep next does not require the invoked skill's own procedure and deliverables

[kyle-sexton]

Problem

skills/repo-sweep/reference/next.md lines 53-55 say "Invoke each skill ... Show the findings." Nothing says the invoked skill's own procedure and output contract bind the step. An operator can invoke the skill, skim its procedure, and summarize, and the step still ticks.

Evidence

melodic-software/.github#153, step prompt-audit (claude-api with prompt-audit). The bundled guide's line 3 says "Execute the steps below in order - do not summarize them back to the user" and names its deliverables (Step 5 report, Step 6 diff). The operator skipped listing the Step 1 inventory, the Step 2 git blame, and the Step 4 signal greps, and summarized in chat. The user had to ask whether the skill actually ran. A redo reached the same result, but only after the user caught it.

Operator error is the primary cause. No script can prove a skill ran in full, so the fix is an instruction.

Fix

Add one sentence to next.md step 3.2 or 3.3: run each skill's procedure in full as the skill states it, produce its deliverables in the form it specifies, and show those deliverables (not a summary) as the findings.

[kyle-sexton]

This was generated by AI during triage.

Triage: verified, briefed (delegable)

Category: enhancement (instruction gap in a skill procedure). Type: Task. priority: medium (assessed default; nothing is blocked on it and there is no deadline).

Verified against main (read-only): the repo-sweep next procedure's "Run the step" section tells the operator to invoke each skill in the entry, then "Show the findings." No sentence says that the invoked skill's own procedure and output contract bind the step. A step can therefore tick after an operator summarizes instead of running the skill as written. No existing text covers this, there is no docs/out-of-scope/ ledger, and none of the other open repo-sweep items duplicates it. #4575, #4533, #4532, and #4537 edit the same procedure file for different reasons, so a worker should expect nearby edits.

Brief (behavioral contract):

• In repo-sweep next, the "Run the step" instructions state that each invoked skill's procedure runs in full, as that skill states it, and that its deliverables are produced in the form the skill specifies.
• The findings shown to the user for review are those deliverables themselves, not a summary of them.
• The change is one sentence in the invoke or show-findings step. Nothing else in the procedure changes, and no script or gate is added: the issue notes that no script can prove a skill ran in full.
• Bump the playbooks plugin version according to the repo's release convention.

Acceptance: a reader of the "Run the step" section can tell that summarizing an invoked skill's output, or skipping its procedure steps, does not complete the step.

Status: status: ready. The autonomous-eligible role label is withheld because the class-assigning step must apply it together with a work-class: label. Proposed class: scoped. The change edits a tracked markdown skill body, which is a runtime surface, so it cannot count as mechanical.