Weekly leaderboard

Overview

Once a week the operator posts an ENG/PRODUCT shipping brief to Slack {{slack_channel}}: what changed, who drove it, a ranked PR leaderboard with streaks and deltas, and a standup thread prompt for async input. Git and Slack are the evidence; the output leads with outcomes but includes hard numbers.

The leaderboard agent is a thin wrapper around this skill, fired by the weekly-leaderboard cron in kortix.yaml.

<window> Cover the last 7 days. Compute the cutoff explicitly:

since=$(date -u -d '7 days ago' +%Y-%m-%d 2>/dev/null || date -u -v-7d +%Y-%m-%d)

If a prior leaderboard post exists in {{slack_channel}}, prefer "since the last post" so nothing is double-counted or skipped — read it first with slack history. </window>

<sources> Pull from real data — never invent contributions.

1. Merged PRs — counts per person (the leaderboard backbone)

Repos: {{target_repos}} (space-separated). Count merged PRs per author in each:

since=$(date -u -d '7 days ago' +%Y-%m-%d 2>/dev/null || date -u -v-7d +%Y-%m-%d)
for repo in {{target_repos}}; do
  gh pr list --repo "$repo" --state merged --search "merged:>=$since" --limit 200 \
    --json number,title,author,mergedAt,labels
done | jq -s 'add | group_by(.author.login)
  | map({author: .[0].author.login, count: length, prs: map({number,title,mergedAt})})'

Exclusions. Drop PRs authored by app/dependabot, app/github-actions, any agent-, and any login ending in [bot]. Count them on a separate line: "+ N bot/agent PRs (dependabot ×N, …)"*. Do NOT exclude a PR a human drove even if an agent helped write it — attribute to the human who opened/merged it.

Identity map (GitHub login → display name). Edit this for your team:

# githubLogin  ->  Display Name
# alice-gh     ->  Alice
# bob-eng      ->  Bob

When in doubt, use the PR author's display-name field.

Also pull commits pushed straight to the default branch (no PR):

for repo in {{target_repos}}; do
  gh api "repos/$repo/commits?since=${since}T00:00:00Z&per_page=100" --paginate \
    --jq '.[] | {sha: .sha[0:8], author: .commit.author.name, msg: (.commit.message | split("\n")[0])}'
done

2. Prior week (deltas + streaks)

Read .kortix/memory/weekly-digests.md — the ### PR Leaderboard and ### Streaks of the most recent entry — for last week's per-person counts (compute +N / -N) and current streak counts.

3. Team discussion (context that makes numbers meaningful)

slack channels --limit 1000
slack history --channel <main-channel-id> --limit 200
slack search --query "shipped after:$since"

Capture launches, decisions, blockers, direction changes, shout-outs. Link the most useful thread. Read last week's standup thread (ts in the ledger).

4. Don't double-count

One logical change = one narrative line. The leaderboard counts raw merged PRs; "What happened" groups them into outcomes. </sources>

<building> A brief a founder can read in under 90 seconds and the team wants to open.

1 — Header ENG/PRODUCT weekly — <one-line takeaway>

2 — Snapshot <n> PRs merged · <n> active shippers · <n> major discussions

3 — What happened (5–7 bullets, ordered by impact): • Area: outcome → why it mattered — Owner(s) (<url|#PR>) Lead with outcomes, collapse fixups into one shipped thing, include Slack-only events. Banned filler: "great", "big week", "kudos", "various improvements".

4 — PR leaderboard (from the per-person counts, sorted desc):

*PR leaderboard — week of YYYY-MM-DD*
🥇 Alice     ██████  6  (+2)
🥈 Bob       ████    4  (=)
🥉 Carol     ███     3  (-1)
   Dave      ██      2  (new)
+ 12 bot/agent PRs (dependabot ×8, agent ×4)

Bar: one █ per PR, max 10 wide (scale past 10). Delta vs last week (+N/-N/=, new if 0 last week). 🥇🥈🥉 for top 3. Only people with ≥1 PR. If nobody merged any, skip this and note it in the snapshot.

5 — Streaks — for anyone shipping ≥1 PR ≥2 weeks running: 🔥 Alice — 8-week streak Read/update counts from the ledger (increment shippers, reset the rest). Only show if at least one streak ≥2.

6 — Standup thread prompt — post as a reply to the main message:

🗓 Monday async standup — reply in this thread by EOD

• What did you ship last week? (link PRs or describe)
• Top priority this week?
• Any blockers?

{{team_mentions}}

7 — Slack signal / next watch — 1–3 bullets for decisions/blockers/asks that didn't map to a PR. </building>

<ledger> After posting, append a ## Week of YYYY-MM-DD entry to .kortix/memory/weekly-digests.md with: a PR Leaderboard table (Person | PRs | Delta | Streak) + a bot/agent line, a Streaks list, the standup thread ts (+ channel id), and short Product / Decisions / Open questions notes. Keep the last 8 weeks; drop the oldest when you exceed it. Land it as a self-merged memory: leaderboard YYYY-WW CR scoped to that file only, AFTER posting to Slack (so a Slack failure doesn't block the ledger). </ledger>

<posting> Post to {{slack_channel}}. Resolve the id and join if needed:

chan=$(slack channels --limit 1000 | grep -i "$(echo '{{slack_channel}}' | tr -d '#')" | awk '{print $1}' | head -1)
slack join --channel "$chan" 2>/dev/null || true

Main post: Block Kit with sections 1–5 + 7. Load kortix-slack for the exact slack send --channel <id> --blocks-file <path> --text <fallback> mechanics. Pass --channel explicitly (cron session, no active Slack turn); always include a plain --text fallback. Standup thread: capture the sent message ts, then slack reply --channel "$chan" --thread <ts> --text "<standup prompt>"; save the ts to the ledger. If nothing shipped, post one honest line + the standup prompt anyway — the team still checks in. </posting>

Guardrails

• Real data only. Every count, delta, and streak traces to gh pr list.
• No secrets / no PII — link PRs; never paste tokens or customer data.
• One main post per run. The standup reply is part of the same run.
• Attribute fairly. Bots are excluded from the count and called out separately; a human-driven, agent-written PR counts for the human.
• Streak integrity. ≥1 merged PR per week, continuous; a missed week resets.