review + fleet-wide: ci-log-auditor's 'add' rationale misdescribes gh --paginate's array merge, plus ~15 sites omitting per_page=100

[kyle-sexton]

#2238 covers ci-log-auditor.md:22 (a missing per_page=100). This is a different defect in the same file — the prose, not the command — and a repo-wide sweep of the same conformance gap #2238 reports at one site.

Part 1 — the add rationale misdescribes what gh actually does

plugins/review/agents/ci-log-auditor.md:65 (mirrored in plugins/review/CHANGELOG.md:20):

Concatenated pages are arrays, so they are combined with add, not by reaching through a wrapper

Measured (gh api --paginate, per_page=10, jq -s 'length'):

array endpoint   (issues/<n>/comments)         -> 1 document
object endpoint  (commits/<sha>/check-runs)    -> 3 documents

gh --paginate merges top-level array responses into a single document; it emits one document per page only for object-envelope responses. So on the /annotations endpoint there is exactly one array, and add is unwrapping a 1-element slurp rather than concatenating pages.

The published command is correct — jq -s '... add | length' yields the right count, verified against output.annotations_count. Only the explanation is wrong. Worth fixing because it is a mechanism claim in a file whose whole subject is being factually right about pagination, and because a reader who believes it will mispredict the shape on the next endpoint.

Part 2 — repo-wide sweep of --paginate without per_page=100

#2238 reports this at one site; it holds at roughly fifteen runnable command sites across four plugins. Stated honestly: --paginate alone is complete — it fetches every page regardless — so these are not truncation defects. They are non-conformant with rule 1 as skills/pull-request/reference/readiness.md:55 publishes it (--paginate with per_page=100), and they cost 3.3x the requests.

Sites on main (c1b4c629), excluding prose mentions, single-resource GETs, and mutations:

file	line
plugins/review/agents/ci-log-auditor.md	22 (already #2238 / #2239)
plugins/source-control/skills/pull-request/reference/merge.md	14, 15, 16
plugins/source-control/skills/pull-request/SKILL.md	166, 167, 168 (C1/C2/C3)
plugins/source-control/skills/pull-request/reference/monitor.md	195
plugins/source-control/scripts/fetch-all-pr-comments.sh	141
plugins/source-control/skills/babysit-loop/reference/telemetry-upsert.md	39
plugins/source-control/skills/babysit-prs/scripts/babysit_gh.py	441
plugins/source-control/skills/babysit-prs/scripts/request_review.py	186
plugins/claude-ops/skills/lanes/scripts/restart-consumer.sh	774
plugins/claude-ops/skills/lanes/scripts/telemetry-upsert.sh	344
plugins/work-items/skills/attend-queue/SKILL.md	136
plugins/work-items/skills/work-loop/reference/telemetry-upsert.md	43
plugins/work-items/tools/work-item-tracker/adapters/github/common.sh	189
plugins/work-items/tools/work-item-tracker/adapters/github/reclaim.sh	46, 50

merge.md:14-16 and SKILL.md:166-168 are worth a callout: they read the same three endpoints that #2163 changed elsewhere in the same skill, and SKILL.md's are 11-17 lines above lines #2163 did give ?per_page=100.

Verify each line before editing — main moves continuously and #2239 is already in flight against the first row.

[kyle-sexton]

Part 2's first row is already closed — re-swept against current main (33f0df5)

#2239 merged at 2026-08-11T23:36:52Z, after this issue was filed. Verified on current main:

plugins/review/agents/ci-log-auditor.md:22
  gh api --paginate ".../actions/runs/<run-id>/jobs?per_page=100" --jq '...'

per_page=100 is present, and the site no longer appears in the sweep. That row is discharged — do not re-fix it. #2238 is likewise closed by #2239.

Part 1 still stands, unchanged. ci-log-auditor.md:65 still reads "Concatenated pages are arrays, so they are combined with add, not by reaching through a wrapper", which misdescribes the mechanism — gh --paginate merges top-level array responses into a single document (measured: array endpoint → 1 document, object endpoint → 3), so add unwraps a one-element slurp rather than concatenating pages. #2239 fixed the command on line 22 and did not touch this prose.

Re-swept Part 2 list, current as of 33f0df5

Runnable REST command sites still carrying --paginate without per_page=100:

file	line
plugins/source-control/skills/pull-request/reference/merge.md	14, 15, 16
plugins/source-control/skills/pull-request/SKILL.md	166, 167, 168 (C1/C2/C3)
plugins/source-control/skills/pull-request/reference/monitor.md	195
plugins/source-control/scripts/fetch-all-pr-comments.sh	141
plugins/source-control/skills/babysit-loop/reference/telemetry-upsert.md	39
plugins/source-control/skills/babysit-prs/scripts/babysit_gh.py	441
plugins/source-control/skills/babysit-prs/scripts/request_review.py	186
plugins/claude-ops/skills/lanes/scripts/restart-consumer.sh	774
plugins/claude-ops/skills/lanes/scripts/telemetry-upsert.sh	344
plugins/work-items/skills/attend-queue/SKILL.md	136
plugins/work-items/skills/work-loop/reference/telemetry-upsert.md	43
plugins/work-items/tools/work-item-tracker/adapters/github/common.sh	189
plugins/work-items/tools/work-item-tracker/adapters/github/reclaim.sh	46, 50

18 sites, down from the original list once ci-log-auditor.md:22 is removed.

Correctly excluded on re-check, so nobody re-adds them:

• reference/review-discipline.md:227 and :237 — continuation lines; the actual commands on the surrounding lines do carry per_page=100.
• claude-ops/skills/morning-brief/scripts/morning-brief.sh:508 and work-item-tracker/adapters/github/README.md:205 — gh api graphql --paginate, which is cursor-paginated with first: 100 in the query. per_page does not apply.
• Test fixtures (morning-brief.test.sh:109, fetch-annotations.test.sh:139, lease-coordination.test.sh:47), CHANGELOG prose, and prompts/loops/loop-lane-prompts.md:190.

Unchanged from the original filing: --paginate alone fetches every page, so none of these is a truncation defect. They are non-conformant with rule 1 as skills/pull-request/reference/readiness.md:55 publishes it, and cost 3.3x the requests. Verify each line before editing — main moves continuously.

[kyle-sexton]

Fully resolved on main — no residual work for a new PR.

Part 1 (ci-log-auditor add rationale): Fixed in #2263 — plugins/review/agents/ci-log-auditor.md now correctly describes that gh --paginate merges array-shaped pages into one document (so jq -s + add unwraps rather than concatenates on /annotations).

Part 2 (--paginate without per_page=100): Fixed across the fleet in #2262 (source-control) and #2263 (review, work-items, claude-ops). Verified on current main: every runnable REST list read in the issue's sweep table now carries per_page=100; babysit_gh.py / request_review.py already had it in the endpoint URL.

Closing as complete.