playbooks: repo-sweep next records every skill as run when an entry does not apply

[kyle-sexton]

Problem

reference/next.md section 2 reads an entry's applies-when but never tells the session to test it before invoking the entry's skills. When an entry does not apply, the only way to close it is tick.sh <id> no-findings <skill@version>..., which takes every skill in the entry.

Observed on melodic-software/.github PR #153, step file-names (playbooks 0.13.0, docs-hygiene 0.23.1):

• applies-when is "repo has a docs tree and a file-name casing rule". The repo has no docs/ tree; inventory.sh printed SCANNED 0 docs.
• docs-hygiene:realign-file-names was never invoked, yet the checklist now reads docs-hygiene:audit-file-names@0.23.1, docs-hygiene:realign-file-names@0.23.1, no findings.
• history.sh derives run / rerun recommendations from these records, so a later plan treats realign as having run at 0.23.1.
• "Ran and found nothing" and "did not apply" produce the same line; nothing distinguishes them.

Proposed fix

1. In next.md section 2, test applies-when with cheap evidence (tracked-file listing, globs) before invoking any skill.
2. Add tick.sh <id> not-applicable <reason>, which records no skill versions, and teach history.sh to treat it as not run.
3. When an entry stops partway (audit found nothing, fix skill not needed), record only the skills actually invoked.

Acceptance

• A not-applicable entry is ticked without invoking its skills and without Playbook-Step or version claims.
• history.sh does not count a not-applicable tick as a run.

🤖 Generated with Claude Code

https://claude.ai/code/session_01S5b57FoPJkJMe1BDwhExiG

[kyle-sexton]

Second occurrence on melodic-software/.github#153, step skill-quality (playbooks 0.13.0):

• applies-when is "repo has skills". The repo has no .claude/skills directory and no tracked SKILL.md.
• The step closed with tick.sh skill-quality no-findings skill-quality:check@0.24.10, so the checklist records the entry as run.
• The next entry, evals-validate, has the same applies-when and will hit the same path.
• The PR body has no Not run: list, so plan left both skill-dependent entries checked in a repo with no skills.

[kyle-sexton]

This was generated by AI during triage.

Partly fixed by #4863: next re-checks applies-when before each step and ticks not-applicable "<evidence>", and history skips those lines. Kept open for the residual below.

Agent Brief

Type: Bug
Summary: A repo-sweep entry whose applies-when is false records no skill as run.

Current behavior:
The second occurrence (a repo with no skills) shows skill-quality ticked no-findings skill-quality:check@..., so the checklist records the entry as run; evals-validate has the same applies-when and hits the same path.

Desired behavior:
An entry whose applies-when is false ticks not-applicable with evidence and records no skill@version. The plan's Not run: list names skipped skill-dependent entries.

Acceptance criteria:

• In a repo with no SKILL.md files, skill-quality and evals-validate tick not-applicable with evidence and list no skill@version.
• A regression test in the repo-sweep script tests covers it.

Out of scope:

• Recording for an entry that stops partway, and the instruction-placement decline path (playbooks repo-sweep: instruction-placement entry has no path to persist declines or record a skipped realign #5167).