tifandotme Recapping standup
The audience is the daily standup: teammates who know the products but not the code.
1. Resolve the window
yesterdayon a Monday means the previous Friday.last weekmeans Monday through Friday of the previous calendar week.
Do not ask the user to confirm the dates.
2. Gather
gh search prs --author=@me --updated=<start>..<end> --json repository,title,url,state,updatedAt,isDraft --limit 100
Merged PRs are the record of what shipped. Open and draft PRs are work in progress, and their timestamps go stale — before reporting one as active work, check that it really moved inside the window.
For every PR returned, fetch its description before writing any bullet:
gh pr view <url> --json body
The title names the change; the body carries the defect, the symptom, and the reasoning a title can't hold. Write from the body, not the title.
Slack, via slack_search_public_and_private, sort: timestamp, include_context: false:
- one day:
from:me on:YYYY-MM-DD - a range:
from:me after:<day before start> before:<day after end>
Slack carries the why behind each PR cluster, plus the work that produced no commit: debugging help, reviews, thread decisions.
3. Write
Group by workstream, not by repository and not by chronology. One bullet per workstream, ordered by how much the standup audience cares.
Each bullet names what changed for a person: a user, a teammate, an environment. Repository names, PR titles, PR numbers, framework names, and file paths stay out. Product names, dataset names, feature names, teammate names, and real numbers stay in — they are what makes a bullet specific.
Separate what shipped from what is still in review.
Drill down on every bullet. Name the actual defect and the symptom someone saw:
Fixed a data import bug.
becomes
Fixed an import bug where all images from the PAS9 labeling batch were silently skipped, because they lived in an archive storage location our import tool wasn't checking.
and
Helped a teammate with an analytics query.
becomes
Helped Novi figure out why her PAS9 dataset image counts didn't match up (20.5k uploaded vs. 16.5k in her report) — her query was only counting images that already had annotations attached, not the full upload set.
Output is a bold window heading and bullets. No preamble, no closing summary.