Daily Code Review

Daily Code Review Report

No em-dashes. Nothing this skill writes may contain an em-dash; use a comma, colon, or parentheses instead. Enforced by a repo hook that flags em-dashes in generated .md. See /flagrare:write-docs.

Generate a team-wide pull request status report focused on actionable next steps. The report surfaces stale PRs, items needing the runner's personal attention, and a quick FYI on active work, so the reader knows exactly what to do when they open GitHub.

Setup (first run only)

Team configs live at ~/.claude/skills/flagrare/daily-code-review/teams/*.json: one file per team, outside the plugin tree so they survive plugin updates and reinstalls.

In Bash, expand ~ explicitly: "$HOME/.claude/skills/flagrare/daily-code-review/teams". The directory may not exist yet, mkdir -p "$HOME/.claude/skills/flagrare/daily-code-review/teams" before writing.

Step 1: Migrate legacy configs (one-time)

If the new dir has no team files but the legacy per-plugin dir does, migrate them:

LEGACY="{skill_directory}/teams"   # old location, lost on plugin reinstall
NEW="$HOME/.claude/skills/flagrare/daily-code-review/teams"
if [ -d "$LEGACY" ] && [ ! -d "$NEW" ]; then
  mkdir -p "$NEW"
  cp "$LEGACY"/*.json "$NEW"/ 2>/dev/null || true
fi

Tell the user once: "Migrated your team configs from the old per-plugin location to ~/.claude/skills/flagrare/daily-code-review/teams/ so they survive plugin updates."

Step 2: Check for existing configs

Check for config files matching ~/.claude/skills/flagrare/daily-code-review/teams/*.json. If none exist, run the first-time setup flow.

First-time setup

Use AskUserQuestion to collect:

1. GitHub org: the GitHub organization to search (e.g., acme-corp)
2. Team name: a human label for the report header (e.g., Platform Team)
3. Team members: for each person, their GitHub login and display name. Ask in a single prompt, one member per line, format: github_login / Display Name

Save to ~/.claude/skills/flagrare/daily-code-review/teams/{team-name-slug}.json:

{
  "org": "acme-corp",
  "team_name": "Platform Team",
  "members": [
    { "github_login": "aturing", "display_name": "Alan Turing" },
    { "github_login": "ghopper", "display_name": "Grace Hopper" },
    { "github_login": "dknuth", "display_name": "Don Knuth" }
  ]
}

Confirm the config with the user before proceeding.

Returning user

If exactly one team config exists, use it. If multiple exist, ask which team to report on.

If the user says "add a team" or "edit team," update or create the relevant config file and re-confirm.

Detect "Me"

Run:

gh api user --jq '.login'

Match against the team members list by github_login. If no match, ask the user which member they are, they might be authenticated with a personal account that differs from their team login.

Data Collection

Use only these GitHub API endpoints. The comments API is noisy and mergeable_state is unreliable, skip both. Staleness comes from updated_at alone.

Open PRs per member

The search/issues endpoint does NOT return draft status reliably (it comes back null). Run two searches per member to separate drafts from non-drafts:

gh api "search/issues?q=org:{org}+is:pr+is:open+-is:draft+author:{login}&per_page=100" \
  --jq '.items[] | {number, title, html_url, updated_at, draft: false, repo: (.repository_url | split("/") | last), user: .user.login}'

gh api "search/issues?q=org:{org}+is:pr+is:open+is:draft+author:{login}&per_page=100" \
  --jq '.items[] | {number, title, html_url, updated_at, draft: true, repo: (.repository_url | split("/") | last), user: .user.login}'

Run all searches in parallel across team members. Deduplicate by PR number + repo.

Reviews and requested reviewers

For each PR:

gh api "repos/{org}/{repo}/pulls/{number}/reviews" \
  --jq '[.[] | select(.user.login | test("\\[bot\\]$") | not) | {user: .user.login, state}]'

gh api "repos/{org}/{repo}/pulls/{number}/requested_reviewers" \
  --jq '{users: [.users[].login], teams: [.teams[].slug]}'

Filter out bot reviews (logins ending in [bot]), they're noise from CI integrations, not human review activity.

Run these in parallel across PRs where possible. If you hit rate limits, back off and retry.

Classification

Staleness thresholds

Calculate hours (or days for parked drafts) since updated_at relative to now. Round to the nearest whole number.

Review state

Determine per-PR by reading the reviews list chronologically:

• Approved: at least one APPROVED review, no subsequent CHANGES_REQUESTED
• Changes requested: most recent non-dismissed review is CHANGES_REQUESTED
• Pending: has requested reviewers who haven't submitted a review

Show reviewer names in each state (e.g., "approved by Alice, Bob").

Report Format

# {team_name} Code Review Report: {YYYY-MM-DD}
> Generated for **{display_name}** | {n} open PRs across {m} members

Section 1: Stale PRs (>24h no action)

All open PRs across the team (including drafts) where updated_at > 24h. Only exclude drafts older than 30 days, those go in the "Parked Drafts" section instead. Label draft PRs with a [Draft] tag so they're visually distinct.

For each PR, show:

Sort by staleness descending. "Next action" should be specific and direct: "Merge, already approved," "Address feedback from Carol," "Needs a reviewer assigned."

If no stale PRs: "No stale PRs, the team is on top of reviews."

Section 2: Needs My Attention (>12h no action)

Two sub-sections:

PRs I need to review PRs where I'm a requested reviewer and haven't submitted a review yet. Show: author, title+URL, hours waiting.

My PRs needing action My own PRs that are approved (merge them!) or have changes requested (address feedback). Show: title+URL, state, hours since last update.

If nothing in either sub-section: "Nothing needs your attention right now."

If the user has stale PRs in Section 1 but none qualify here (e.g., all are drafts or pending first review), add a brief note pointing them back: "Your 3 open PRs are stale but waiting on reviewers or still in draft, see Section 1 above." This bridges the gap so Section 2 doesn't feel disconnected when all the user's action items are upstream.

Active (FYI)

My open PRs or pending review requests updated within the last 12 hours. One line each: title, URL, current state. No action needed, just awareness.

If nothing active: omit this section entirely.

Parked Drafts

Draft PRs from any team member untouched for 30+ days. Show: owner, title+URL, days since update. These are informational, the team may want to close or revive them.

If none: omit this section.

Output

Always render the full report as formatted markdown in the conversation.

If the user asks to save to a file, write to the path they specify. If they say "save it" without a path, default to ./pr-report-{YYYY-MM-DD}.md.

Edge cases

• A PR can appear in both Section 1 (pod-wide stale) and Section 2 (needs my attention). That's expected, different audiences for the same PR.
• If a team member's PRs can't be fetched (permissions, API error), note it in the report footer and continue with the rest.
• If gh is not installed or not authenticated, tell the user to run gh auth login first.