pr-activity-report Skill

pr-activity-report

Arguments: $ARGUMENTS (required). Format: owner/repo optionally followed by a quarter like Q1 2026. If no quarter given, use the current quarter.

Quarter date ranges

Q1: Jan 1 - Mar 31 (start data from the Monday of the week containing Jan 1)
Q2: Apr 1 - Jun 30
Q3: Jul 1 - Sep 30
Q4: Oct 1 - Dec 31

Step 1: Collect data via `gh` CLI

You need three datasets. Exclude bot authors (login containing [bot] or dependabot or renovate or github-actions). Only count merged PRs.

Dataset 1: Weekly authored/reviewed counts

For every human contributor with repo access, compute per-ISO-week:

prs_authored: number of PRs they authored that were merged that week
prs_reviewed: number of unique PRs they submitted a review on that week

Use gh api with GraphQL or REST pagination. Group by ISO week (Monday start). Include weeks with zero activity so the heatmap has no gaps.

Collect all users who either authored or reviewed at least one PR in the period, plus anyone with push access who had zero activity (to show gaps).

The output shape is:

[
  { "week_start": "2026-01-05", "user": "alice", "prs_authored": 3, "prs_reviewed": 5 },
  ...
]

Dataset 2: Review pairs

For every (reviewer, author) pair where the reviewer submitted at least one review on the author's merged PRs during the quarter:

[
  { "reviewer": "alice", "author": "bob", "count": 12 },
  ...
]

Dataset 3: Per-PR detail

For every merged PR in the quarter, collect size and classification data:

[
  {
    "number": 42,
    "user": "alice",
    "title": "feat(auth): add SSO login",
    "merged_at": "2026-01-15T10:00:00Z",
    "additions": 340,
    "deletions": 12,
    "changed_files": 8,
    "size_tier": "L",
    "pr_type": "feat",
    "top_dirs": ["src/auth", "tests/auth"]
  },
  ...
]

Size tiers (based on additions + deletions):

| Tier | Lines changed | |------|--------------| | XS | 0-9 | | S | 10-49 | | M | 50-249 | | L | 250-999 | | XL | 1000+ |

Top directories derived from the file paths (first two path segments, deduped, max 5).

Data collection approach

Use gh api with pagination. Suggested approach:

# Get all merged PRs in date range, including size fields
gh api --paginate "repos/{owner}/{repo}/pulls?state=closed&sort=updated&direction=desc&per_page=100" \
  --jq '.[] | select(.merged_at != null) | select(.merged_at >= "START" and .merged_at < "END") | {number, user: .user.login, title: .title, merged_at: .merged_at, additions, deletions, changed_files}'

# For each PR, get reviews
gh api "repos/{owner}/{repo}/pulls/{number}/reviews" \
  --jq '[.[] | .user.login] | unique'

# For each PR, get file paths (batch alongside reviews)
gh api "repos/{owner}/{repo}/pulls/{number}/files" \
  --jq '[.[] | .filename]'

From the file paths, derive top_dirs by extracting the first two path segments of each file (e.g. src/auth/login.ts -> src/auth), deduping, and keeping the top 5 by frequency.

Compute size_tier from additions + deletions using the tier table above.

Parse pr_type from title using the regex above.

This will be slow for large repos. To speed up, batch requests and cache results. If there are more than ~500 PRs, process in chunks and show progress.

If the API calls are going to be very numerous (>200 PRs), write a small bash script that collects everything into a JSON file first, then process it.

Step 2: Compute summary stats

From the collected data, compute:

Total human PRs merged in the quarter
Active contributors: count of people with >= 1 authored or reviewed PR
Top author share: what % of total PRs the #1 author accounts for, and their count
Review coverage: how many people handle 50%+ of all reviews (sort by review count desc, cumulative sum until > 50% of total reviews)
Median PR size: median of (additions + deletions) across all PRs, with its size tier label (e.g. "142 lines (M)")
Code volume: total additions and deletions for the quarter, formatted as +N / -N (use k suffix for thousands, e.g. +2.1k / -0.4k)

Display all six stats as cards in a 3x2 grid.

Step 3: Write insights

Analyze the data and write opinionated, specific insights. Reference actual numbers and usernames. Cover:

Key Takeaways (top-level)

3-5 bullet points covering the most important patterns:

Authorship concentration (is one person doing disproportionate work?)
Review load distribution (top-heavy? evenly spread?)
Author/review imbalances (people who author a lot but rarely review, or vice versa)
Size distribution skew (are most PRs XS/S suggesting good decomposition, or lots of XL suggesting review bottleneck risk?)
Codebase hotspots (top 3-5 directories by PR count, are changes concentrated or spread?)
Any notable patterns (spikes, gaps, siloed clusters)

Per-section insights

Write a short insight box before each of the three heatmaps, the chord diagram, the size distribution chart, and the contributor profiles section:

Authored PRs: outlier weeks, consistent vs intermittent contributors
Review Activity: who leads, broadly distributed vs narrow reviewers, people with few reviews
Author/Review Balance: healthy ratios, persistent imbalances, caveats
Review Network: bus factor, reciprocity, siloed clusters, recommendations
Size Distribution: who ships large PRs vs small ones, team-wide size habits, outliers
Contributor Profiles: notable specialization patterns, who touches the broadest vs narrowest parts of the codebase

Be direct. Use color-coded markers: .good (green), .flag (yellow), .concern (red).

Add a context disclaimer noting this data covers only this repo and contributors may be active elsewhere.

Step 4: Generate HTML

Generate a single self-contained index.html file in a temp directory (mktemp -d). Use Plotly 2.35.2 and D3 v7. Dark GitHub theme.

The HTML must include, in this order:

Title: {Quarter} {Year} -- PR Activity (Humans Only) with subtitle
Summary stat cards in a 3x2 grid (6 cards total)
Context disclaimer insight box
Key takeaways insight box
Filter buttons (Active contributors >= 3, All with any activity, Everyone)
For each of the three heatmaps (Authored, Reviewed, Ratio):
- Insight box with analysis
- Panel with Plotly heatmap
Size Distribution section (stacked bar chart)
PR Type Breakdown section (conditional, stacked bar chart)
Chord diagram section:
- Insight box with review network analysis
- Threshold filter buttons (>= 5, >= 3, All)
- D3 chord diagram in SVG with viewBox scaling
Contributor profile modal (hidden, shown on click)

Embed the three JSON datasets directly in <script> tags.

Heatmap details

Authored & Reviewed: Red-green activity scale using 90th percentile for zmax. Numbers displayed in cells. Sorted by total reviews descending.
Ratio: Green (net reviewer) -> gray (balanced at 1:1) -> red (mostly authoring). Scale 0-5, infinity symbol for authored-with-zero-reviews. Show colorbar.
Y-axis labels: username (A:NNN R:NNN +Nk -Nk) with monospace font, padded. The +Nk -Nk shows total additions/deletions for that contributor, using k suffix for thousands (e.g. +2.1k -0.4k). For values under 1000, show the raw number (e.g. +340 -12).
All heatmaps: 32px cell height, 2px gap, dark background, no mode bar

Size Distribution chart (new)

Horizontal stacked bar chart (Plotly) showing each contributor's PR count broken down by size tier. One bar per contributor (same Y-axis order as heatmaps).

Colors for size tiers:

XS: #7d8590 (gray)
S: #58a6ff (blue)
M: #3fb950 (green)
L: #d29922 (orange)
XL: #f85149 (red)

Same filter buttons apply. Place after the three heatmaps, before the chord diagram.

PR Type Breakdown chart (conditional, new)

Only shown when >= 50% of PRs in the quarter have a parseable conventional commit prefix in the title (matching the regex from Dataset 3).

When shown: horizontal stacked bar per contributor with colors per type. Use a distinct color palette (not the size tier colors). Same filter buttons apply.

When omitted: add a brief note in the takeaways explaining that PR type breakdown was skipped because fewer than 50% of PR titles follow conventional commit format, and include the actual parse rate percentage.

Chord diagram details

Colors from a 25-color palette
Sorted by total involvement (reviews given + received)
Hover on arc: show who they review for and who reviews them
Hover on ribbon: show bidirectional review counts
Tooltip with fixed positioning, dark background
viewBox-based SVG so it scales responsively

Contributor Profile modal (new)

Clicking any username in any Y-axis label, chart label, or stacked bar opens a modal with:

Header: username, GitHub avatar (from https://github.com/{username}.png?size=64), total authored and reviewed counts
Top PRs by size: top 5 PRs by lines changed (additions + deletions), each showing title (linked to the PR on GitHub), size tier badge with tier color, +N -N, and merge date
Size tier distribution: small donut chart (D3, already loaded) showing their personal size tier breakdown
Codebase areas: horizontal bar chart of top 5 directories they touched (by PR count), using the top_dirs from Dataset 3

Implementation: pure CSS modal toggled by JS. Fixed overlay with --panel background card, max-width 600px, scrollable content. Close on overlay click or Escape key. No new libraries.

CSS theme

--bg: #0d1117; --fg: #e6edf3; --muted: #7d8590;
--panel: #161b22; --border: #30363d; --accent: #58a6ff;

Size tier color variables (for reuse in charts and badges):

--tier-xs: #7d8590; --tier-s: #58a6ff; --tier-m: #3fb950;
--tier-l: #d29922; --tier-xl: #f85149;

Step 5: Output

Print the path to the generated index.html file. The user can then deploy it however they choose (e.g. surge, gh-pages, local preview).

Important notes

If the repo has fewer than 10 PRs in the quarter, mention that in the summary and skip the chord diagram.
Handle GitHub API rate limits gracefully. If you hit limits, use conditional requests or wait.
The entire output is a single HTML file with no external dependencies beyond the two CDN scripts.
Do NOT anonymize usernames. Use actual GitHub logins.
Sort everything consistently: by total reviews descending, then authored descending, then alphabetical.

Agent Skills: pr-activity-report

Install this agent skill to your local

Skill Files