_ __    ___ ___    _ __   ___  
| '__|  / _ ` _ \  | '_ \ / _ \ 
| |    | (_|  __/  | |_) | (_) |
|_|     \__,_\___| | .__/ \___/ 
                   |_|          

raepo

A command-line tool for understanding how long pull requests take to merge in a GitHub repository — broken down by author, with honest statistics.

$ raepo karpathy/nanochat --since 2026-01-01 --sort tail

fetched 154 PRs.
Repo:    karpathy/nanochat
Window:  2026-01-01 → 2026-05-04
PRs:     154 analyzed (28 merged, 126 closed-unmerged)

Author           merged  typical  average   tail  accept
────────────────────────────────────────────────────────
georgeshakan          1     2.4h     2.4h   2.4h    100%
Yamahammer            1     4.5h     4.5h   4.5h    100%
dipeshbabu            1     8.4h     8.4h   8.4h     33%
adriablancafort       1    21.7h    21.7h  21.7h    100%
...
svlandeg             11     1.4d     3.6d   5.9d     85%
aarushisingh04        1     7.2d     7.2d   7.2d    100%
marcinbogdanski       1    10.1d    10.1d  10.1d    100%
BlackSamorez          1    10.2d    10.2d  10.2d    100%
why2011btv            1    11.3d    11.3d  11.3d    100%
mathieu-lacage        1    12.4d    12.4d  12.4d    100%
2bitbit               1    13.5d    13.5d  13.5d    100%
ykirpichev            1    14.3d    14.3d  14.3d     50%
Jah-yee               1    15.9d    15.9d  15.9d     14%

No accepted PRs (81):
  aidev2o25, ACautomata, manmohan659, zolopgh, sranganath2,
  faitholopade-source, jake-molnia, giovannizinzi, hiSandog,
  dengdx, RohanKhanBD, hbfreed, JoseRodriguez26,

Install

Tool	Command
One-off (no install)	npx raepo … / bunx raepo …
Global (npm)	npm i -g raepo
Global (bun)	bun add -g raepo

After install, both raepo and ræpo resolve to the same binary. (npm package names are ASCII-only, so the published package is raepo; ræpo is wired up via the bin field for shell use.)

Usage

raepo <org>/<repo> [options]
raepo config <get|set|list> [args]

Filters
  --since <date>          Only PRs created on or after <date> (YYYY-MM-DD)
  --max-age <duration>    Only PRs created within <duration> (e.g. 30d, 12w)
  --author <list>         Restrict to author(s); comma-separated, repeatable
  --include-bots          Include bot accounts (excluded by default)

Output
  --format <f>            plain | json | csv  (default: plain)
  --sort <key>            merged | typical | average | tail | accept
                          (default: merged, descending)
  --limit <n>             Show top n authors

Auth
  --token <pat>           GitHub personal access token to read private repos.

Other
  --verbose, -v           Print per-request rate-limit info to stderr
  --version, -V           Show version

To get a GitHub token you can run gh auth token (if the gh CLI is installed and authenticated)

What the numbers mean (and don't)

Per-author merge time is a function of the whole system — reviewer availability, CI flakiness, PR size, time zones — not just author behaviour. Read this tool as descriptive, not evaluative. Use it to spot bottlenecks (long tails, low acceptance), not to rank people.

Bots (dependabot, renovate, *[bot], anything with user.type === "Bot") are excluded by default because they skew everything; opt in with --include-bots.

Features

What the tool does.

Targets & access

Single repository per invocation: raepo <org>/<repo>. Public repos work without auth (60 req/hour). If you supply a token, you can access private repositories and get more requests to GitHub. On Unix you can pipe it in:

bun run raepo karpathy/nanochat --since 2026-01-01 --token gh auth token

Since a token is so easy to generate, I didn't want the security concern of supporting to save them with raepo config.

Per-author statistics

Raepo reports five stats:

• merged — count of merged PRs
• typical — median time-to-merge (p50)
• average — mean time-to-merge
• tail — 90th percentile
• accept — merged / (merged + closed-unmerged)

Why median and mean: PR merge times are heavily skewed — one 6-month-old PR can quintuple the mean. Median is the honest centre; the mean stays for people who explicitly ask for it. tail captures the long-tail behaviour the median hides; accept shows whether an author's PRs typically land at all. Together these read the shape of the distribution, not just one number from it.

Rejected (won't add later either): per-author "review wait" / "author wait" responsiveness. The signal is too noisy at the per-PR level and the per-author cut reads as a performance metric. The honest "where do PRs get stuck?" view belongs in a repo-aggregate bottleneck view (see PLAN.md), not as an author column.

Bots are excluded by default and merged === 0 authors get demoted to a compact "No accepted PRs" line below the table — see Modes and What the numbers mean above.

Only authors filter

The regular pull requests API doesn't support filtering by author, so since some repos might have thousands, we use a different API to get to the PRs when you just want the data for a few authors, so we can fetch data just for those.

• Multiple --author — --author alice,bob or --author alice --author bob. Table includes every specified author, no demotion. Server-side filtering via GitHub's search API, so we don't pull the whole repo.
• Single --author — shows vertical detail view (label:value) for that one author. Again using a serverside author query filter.

Why server-side filtering: a 5,000-PR repo would otherwise burn the rate limit before producing one row. The endpoint switch is encapsulated inside the GitHub provider; the Provider interface stays clean for a future GitLab implementation.

Filters

max-age and since filters are supported. We turn max-age into a since filter before getting pages from the GitHub API, where we stop fetching once we get past the since timestamp.

Defaults & config

raepo reads defaults from ~/.raepo/config.json. Manage it with the config subcommand:

raepo config set format json     # always emit JSON
raepo config set sort typical    # rank by median merge time
raepo config get format
raepo config list                # show all keys with their source (env | config | default)

Schema keys (all optional): format, sort, include-bots, concurrency. The matching env vars are RAEPO_FORMAT, RAEPO_SORT, RAEPO_INCLUDE_BOTS, RAEPO_CONCURRENCY.

Precedence is CLI flag → RAEPO_* env var → config file → built-in default. Tokens are deliberately not a config key — use $GITHUB_TOKEN or gh auth token so they don't sit on disk.

Exit codes

Code	Meaning
0	Success (or zero matches for a queried author — that's not an error)
1	User error: bad flag, malformed <org>/<repo>, invalid config key/value, malformed config file
2	Network or GitHub error: rate limit, 401/404/5xx after retries exhausted

Beyond the brief

• Three output formats — plain (default), json (stable versioned schema), csv. JSON is shaped for jq or for feeding a dashboard.
• View modes that adapt to author count — zero authors = ranked table with demotions; one author = vertical label:value detail; two or more = table without demotion. All three formats stay available via --format.
• config subcommand — persisted defaults at ~/.raepo/config.json with a versioned schema and RAEPO_* env-var overrides.
• --verbose mode — surfaces per-request x-ratelimit-remaining / reset to stderr; useful when planning a long run against a private repo.
• provider — GitHub API is implemented as provider so we can add support for GitLab and other hosts in the future.
• simple code structure - for a simple codebase. Trying to separate in single concern files.
• TDD - Tests were generated by Claude. They seem good, but have not been vetted in detail. This is a known technical debt.
• Minimal CI - GitHub Actions to run all tests. If green we can publish to npm, but npm login hasn't been set up yet.

Production-readiness signals

The brief asked for these to be "clearly recognizable" without full implementation. The choices:

Concern	Approach
Network failures	5xx retries with 1s/2s/4s exponential backoff (max 3); 4xx and rate-limit responses not retried (different signal)
Rate limits	Typed ProviderRateLimitError with reset-time hint; --verbose surfaces remaining quota per request
Configuration	Versioned schema (version: 1) with precedence CLI > env > file > built-in default; resolver returns per-field source for config list
Future providers	All GitHub specifics behind a Provider interface; provider='github' hardcoded for now, GitLab can drop in via one switch case
Testability	fetch and sleep injected via provider config; pure modules for parse/stats/output/config; ~230 tests including live acceptance tests against karpathy/nanochat
CI + release	Single GitHub Actions workflow runs install/check/typecheck/test/build/smoke; auto-publishes to npm when package.json version is new; --provenance via OIDC
Exit codes	Standard CLI contract — see the table below

Development

Built with Bun + TypeScript.

bun install
bun test
bun run build           # produces dist/raepo.js
bun run raepo …         # run from source

The published npm package ships the bundled JS so it runs on Node too.