Brand voice review
“Microcopy reviewed against the brand guide. Button labels follow verb-noun. Toasts ≤ 80 chars. No exclamation marks outside the success state.”
cat. № YOU-001Clud Bug.
AI PR review with project-aware skills. Install the GitHub App, write a skill, get reviews graded against your conventions on every pull request.— Cluddus bugfindii, observed crawling on every PR.
Install the GitHub App
Managed tiers — see pricing and compare on app.cludbug.dev, or self-host the open-source workflow for free.
Three steps. Two minutes per review.
- 1
Install
Install the GitHub App on the org or repos you want reviewed. Approve the permissions for pull requests, contents, and checks.
- 2
Authorize
Pick a plan on the dashboard — managed tiers for any repo, public or private. The bot reviews against the skills in
.claude/skills/, starting with the baseline kit. - 3
Reviews land
Every PR (and every push to one) gets an inline review within ~2 minutes — comments cited by skill name, anchored at the exact line.
Skills are how Clud Bug knows your codebase.
Generic PR review tools evaluate your code against generic best practices. Clud Bug evaluates it against yourstandards — encoded as plain Markdown the bot loads on every PR. A few of the high-value patterns teams write:
API contract enforcement
“Anything under /v1/* is frozen. Schema changes need a /v2alongside. Flag breaking changes; require deprecation headers on removals.”
Compliance & PII
“No PII (email, phone, name) in logs, ever. No console.log in app/api/*. Every secret read needs an audit log entry.”
Test discipline
“Every new endpoint ships a happy-path and a 4xx test in the same PR. Refactors can’t reduce test count without an explicit note.”
cat. № YOU-004When one pair of eyes isn’t enough.
Most reviews are a single fast pass — skill-aware and cited. On the Team tier, Clud Bug runs a two-pass cross-check: a second naturalist re-reads the first one’s findings against the diff, and a third — the arbiter — is called in only when the two disagree:
Beetle
The broad scan. Walks the full diff, surfaces every candidate issue, no filtering. Optimized for recall.
Wasp
The cross-check. Re-reads Beetle’s findings against the diff and the skills, drops noise, escalates the real ones, and catches what Beetle missed.
Mantis
The arbiter. Only fires on disputes — where Beetle and Wasp disagree on severity or correctness. Returns a single decisive call with reasoning.
Three perspectives, one PR thread, citations all the way down. Compare tiers →
From Clud Bug’s notebook.
This round: 0 critical · 2 minor · 0 resolved from prior · 0 still open
Even the largest specimens get a full examination.
Every PR review gets a budget tailored to its diff. Clud Bug looks at the changed lines first — a one-file typo gets a quick pass; a thousand-line refactor gets the time it needs. The bot is told its own budget upfront and checks in with itself mid-review, so the summary at the bottom of your PR is always a finished thought, never a half-written sentence.
Why this matters.Stock PR-review tools pick a fixed turn count and hope. On small PRs they overspend; on large PRs they run out and leave you with an incomplete review. The budget-aware approach: tiny PR gets ~5 turns of attention, large PR gets ~25, very large gets ~40 — same Clud Bug, same quality bar.
Prefer your own runner? Ship the workflow.
The hosted GitHub App is the recommended path — managed runner, managed billing, no secrets to rotate. For air-gapped orgs and teams that want to bring their own Anthropic key, the same review engine ships as an open-source npm package and runs as a GitHub Action under your own credentials.
npx clud-bug init 🐛 Field season opens here. baseline kit: 4 specimens pinned 4 to .claude/skills/ wrote .github/workflows/clud-bug-review.yml git add .claude .github/workflows/ git commit -m "Add clud-bug" && git push