[haacked]

GitHub's search API returns 30 results per page unless the call asks for more. The standup skill ran its three searches as single gh api search/issues calls, so a standup window with more than 30 merged PRs dropped the rest with no error. On Friday's standup, 14 of 44 merged PRs were missing, including 8 that had carried over from the previous standup.

The three searches now go through scripts/standup-search.sh, which reads every page at 100 per page. It exits non-zero when GitHub returns fewer items than total_count or marks the search incomplete_results. On a failed search, the skill reports the error and does not write the archive file. The next run takes that file's modified time as its merge cutoff, so writing it after a short search would drop the missing PRs for good. bin/standup-run carries the same exception, because its prompt otherwise always writes the archive file.

Test plan

• ai/skills/standup/scripts/tests/test-standup-search.sh passes. It runs against a fake gh that pages results like GitHub. It covers a three-page read, a short read, a match over 1,000 results, an incomplete search, and a failed gh call.
• Each of these breaks fails the suite: removing --paginate, removing per_page=100, or removing the truncation check.
• Live searches for Friday's window returned 45 of 45, 17 of 17, and 11 of 11 matches.
• The next scheduled standup DM includes every merged PR.

https://claude.ai/code/session_01A512d6pECY8WFZS47mRFEY