Contributor Leaderboard

<!-- autoresearch: variation A — better inputs (compare API + PR reviews + first-timer signal), folds in B's "Movement This Week" lede -->

> **${var}** — Target repo to scan contributors of (e.g. "owner/aeon"). If empty, reads from memory/watched-repos.md.

Today is ${today}. Rank the humans behind the fork fleet — who's pushing commits into their forks, who's sending work back upstream, who's reviewing other people's code, and who's building new skills that upstream hasn't seen yet.

This complements `skill-leaderboard` (what is popular) and `fork-fleet` (which forks diverge). This skill asks: **who are the people?**

## Why this exists

The `tweet-allocator` skill rewards social mentions with $AEON. Code contributors get nothing — no recognition, no signal that upstream values their work. This leaderboard is the contributor-side mirror: public recognition for the people actively moving the project forward. Run it weekly, name names, and the flywheel closes.

## Steps

1. **Determine the target repo.** If `${var}` is set, use that. Otherwise read `memory/watched-repos.md` and use the first entry. Store as `TARGET_REPO`. Resolve the upstream default branch once: `UPSTREAM_BRANCH=$(gh api repos/${TARGET_REPO} --jq .default_branch)`.

2. **Fetch all active forks** (pushed within the last 30 days):
   ```bash
   CUTOFF=$(date -u -d "30 days ago" +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -v-30d +%Y-%m-%dT%H:%M:%SZ)
   gh api repos/${TARGET_REPO}/forks --paginate \
     --jq "[.[] | select(.pushed_at > \"$CUTOFF\") | {owner: .owner.login, full_name: .full_name, default_branch, pushed_at, stargazers_count, created_at}]"
   ```
   If no active forks found, log `FORK_CONTRIBUTOR_LEADERBOARD_NO_FORKS` to `memory/logs/${today}.md` and stop (no notification).

3. **Fetch all upstream PRs** in one paginated call. Keep `author_association` so we can flag first-time contributors without an extra call:
   ```bash
   gh api "repos/${TARGET_REPO}/pulls?state=all&per_page=100" --paginate \
     --jq '[.[] | {number, state, merged_at, user: .user.login, title, created_at, author_association}]'
   ```
   Build a map `{login -> {opened: N, merged: N, first_time: bool, pr_titles: [...]}}` keyed on `.user.login`. Set `first_time: true` if any of their PRs has `author_association == "FIRST_TIME_CONTRIBUTOR"`.
   Skip bots: any login ending in `[bot]`, plus `aaronjmars`, `aeonframework`, `github-actions`.

4. **Fetch all upstream PR review comments** in one paginated call (this is the missing reviewer signal):
   ```bash
   SINCE=$CUTOFF
   gh api "repos/${TARGET_REPO}/pulls/comments?since=${SINCE}&per_page=100" --paginate \
     --jq '[.[] | {user: .user.login, pr_url: .pull_request_url, created_at}]'
   ```
   Build `{login -> review_comments: N}`. Apply the same bot filter. Cap at 20 review comments per contributor (someone who left 200 nit comments on one PR shouldn't dominate).

5. **For each active fork, get authored commit count via the compare endpoint** (one call per fork, returns `ahead_by` plus a commits array with author metadata — replaces the prior per-fork pagination loop):
   ```bash
   gh api "repos/${TARGET_REPO}/compare/${UPSTREAM_BRANCH}...${FORK_OWNER}:${FORK_DEFAULT_BRANCH}" \
     --jq "{ahead_by: .ahead_by, owner_commits: ([.commits[] | select(.author.login == \"${FORK_OWNER}\")] | length)}"
   ```
   Record `owner_commits` (capped at 30 by the scoring formula) and `ahead_by` (used in the article narrative, not scored). The compare endpoint returns up to 250 commits — more than enough; if `ahead_by > 250`, treat owner_commits as a lower bound and note it. If the call returns 404 (deleted fork), 422 (no common ancestor), or 409 (empty repo): record `owner_commits: 0, ahead_by: 0` and continue.

6. **Detect new skills** added by each fork owner. For each active fork, list the contents of `skills/` against their default branch:
   ```bash
   gh api "repos/${FORK_FULL_NAME}/contents/skills?ref=${FORK_DEFAULT_BRANCH}" --jq '[.[] | .name]'
   ```
   Compare against upstream's skill directory names (scan this repo's `skills/` locally). Any skill names in the fork but not upstream count as **new skills**. Cap at 5 per fork to prevent mass-rename gaming. If the fork has no `skills/` dir (404), record `new_skills: []`.

7. **Score each contributor** using this formula:
   - `+10` per merged upstream PR (authored by the contributor)
   - `+5` first-time-contributor bonus (one-time, applies if any of their PRs is `FIRST_TIME_CONTRIBUTOR` — first PRs are the highest-leverage signal)
   - `+3` per opened-but-not-merged upstream PR
   - `+2` per upstream PR review comment they left (capped at 20)
   - `+1` per authored commit to their own fork (capped at 30)
   - `+5` per new skill file detected in their fork (capped at 5)
   - `+1` per star on their fork (was +2 — halved to reduce star-farm gaming; cap at 20 stars)

   Rank all contributors by score descending. A contributor is anyone who either owns an active fork OR has authored an upstream PR in the past 30 days OR has left ≥1 upstream review comment in the past 30 days (union of all three sets).

8. **Compare to last week's leaderboard.** Glob `articles/contributor-leaderboard-*.md` from the last 14 days, pick the most recent, and parse its ranked list (logins + scores) using a tolerant regex on the table rows (`^\| \d+ \| @(\S+) \| (\d+) \|`). If parsing yields zero rows, skip the comparison silently (don't crash). Compute week-over-week rank changes (new entries, rank shifts ≥3, dropouts).

9. **Write the article** to `articles/contributor-leaderboard-${today}.md`. Lead with the narrative — the table is the proof, not the headline:

   ```markdown
   # Fork Contributor Leaderboard — ${today}

   *${N_CONTRIBUTORS} contributors moved ${TARGET_REPO} this week across ${N_FORKS} active forks, ${N_UPSTREAM_PRS} upstream PRs, and ${N_REVIEW_COMMENTS} review comments.*

   ## Movement This Week

   *3–5 short paragraphs telling the story of the week.