find-reviewer

claude-find-reviewer

Claude Code plugin that picks the right PR reviewer by ranking contributors on current load, breadth across the repo, and familiarity with the changed files.

When you type /find-reviewer from a branch with an open PR, it auto-detects the PR, fans out a handful of gh queries in parallel, and returns a ranked markdown table with a one-line top pick.

Why this exists

GitHub doesn't expose a "who's free?" view. Picking a reviewer blindly often means the PR sits for days behind someone's existing queue, or lands on someone who's never touched the code. This plugin synthesizes the answer from signals you already have access to:

Signal	What it tells you
Open review requests (org-wide)	Are they already buried under other PRs to review?
Authored open PRs (org-wide)	How many of their own PRs are still in flight?
Recent reviews in this repo (45d)	Do they actually engage with review work here?
Breadth (top-level dirs touched, 180d)	Are they a generalist who can reason about unfamiliar code?
Familiarity (commits to this PR's files, 180d)	Have they specifically worked on these files?
Public-event recency	Are they around right now?

Scoring is min(load, fitness) — a totally idle reviewer with zero reason to trust them as a fit doesn't get recommended just for being free. Someone overloaded but a perfect fit also doesn't win. You want both.

Install

This repo is a Claude Code plugin marketplace with a single plugin. From inside Claude Code:

/plugin marketplace add thomast8/claude-find-reviewer
/plugin install find-reviewer@claude-find-reviewer

Usage

/find-reviewer              # auto-detects the current branch's PR; falls back to repo overview
/find-reviewer 123          # explicit PR number
/find-reviewer --repo-only  # force a general capacity overview even if a PR is detected
/find-reviewer --refresh    # wipe the cache and re-fetch everything

How it works

The plugin ships two bash helpers and one command markdown file:

• scripts/gather.sh — resolves the target (PR auto-detect, explicit number, or repo-only), builds a candidate pool from the last 30 merged PRs + recent committers, then for each of 4 candidates fans out 6 parallel queries: open review requests, authored PRs, assigned issues, last public event, recent in-repo reviews, and breadth. Emits one JSON blob.
• scripts/familiarity.sh — pure jq pass over a pre-built author-per-file cache. Called once per candidate in PR mode. No live API calls.
• commands/find-reviewer.md — the command Claude loads. Describes the workflow, the exact scoring formula, and the output format.

Commit author resolution is server-side. The plugin uses gh api /repos/.../commits?author=<login> and …?path=<file> so GitHub resolves each commit's author to a login internally. This means it works in private org repos where contributors commit with corporate emails that aren't published on their public GitHub profiles.

Caching

Results are cached under $XDG_CACHE_HOME/find-reviewer/ (default ~/.cache/find-reviewer/):

• Per-candidate signals — TTL 15 min, keyed by (repo, login). Most of the wall-time savings.
• File-authors map — immutable, keyed by (repo, PR head SHA). Reused across every candidate in a run and across reruns on the same PR-commit.
• Candidate pool — TTL 1 hr, keyed by repo.
• Commit email → login — appended as new emails are seen, no TTL.

Cold runs take ~25s (4 candidates × 6 parallel queries each). Warm reruns are ~2s. /find-reviewer --refresh wipes the cache if you want fresh data.

Requirements

• gh (authenticated — gh auth status should pass)
• git
• jq
• bash 3.2+ (works on stock macOS)

All three are already required by the closely-related claude-worktree-toolkit; if you have that installed you already have everything.

License

MIT — see LICENSE.