Statement-based PR review

Encode the rules your reviewers already enforce.

CrossCheck turns natural-language standards into pass/fail checks on every PR — running in our managed cloud, against your codebase.

Start your 14-day trial Sign in

Only takes a few clicks to set up · 14-day Pro trial · No credit card required

feat: add events_session aggregation

+7 −0

000CrossCheck

5 checks

All new tables have a Great Expectations test suite
Data quality
queued
Every public function has a docstring
API conventions
queued
Public API changes are recorded in CHANGELOG.md
API conventions
queued
No secrets, tokens, or .env values in source
Security
queued
Migrations are reversible (down() implemented)
Data quality
queued

The problem

Generic AI reviewers comment on style. Your team's bar is higher.

Repo-specific rules

The standards that actually matter — migration shape, logging conventions, ownership of public APIs — never make it into a generic reviewer's prompt.

Per-statement evidence

Every check passes or fails with a line-numbered reference, not a paragraph of vibes. You can act on a CrossCheck result without reading the diff yourself.

Runs on every PR, automatically

Install the CrossCheck GitHub App once. Every PR is reviewed automatically in our managed cloud — no CI workflow to maintain, no API keys to manage.

How it works

From plain English to a check on the PR.

Three steps. No prompt engineering, no glue code, no per-repo plumbing — reviews run in the CrossCheck-managed cloud.

Write a statement

Describe a rule in plain English. Group statements into Check Groups and attach them to repos.

Reviews run in our cloud

Install the CrossCheck GitHub App. Every PR kicks off a fresh, isolated review; the agent proves each statement true against the diff.

Pass / fail, on the PR

Results post back as a check group: pass, fail, or needs review, with line references and evidence. Optionally auto-approve when everything passes.

Statements

What does “a good PR” mean in your repo?

Whatever it is, you can encode it. Statements are natural-language rules the agent must prove true on every PR.

All new tables have a Great Expectations test suite

Data quality

Every public function has a docstring

API conventions

API changes are recorded in CHANGELOG.md

API conventions

No secrets or tokens committed to source

Security

Migrations are reversible (down() implemented)

Data quality

New endpoints have OpenAPI specs

API conventions

All new tables have a Great Expectations test suite

Data quality

Every public function has a docstring

API conventions

API changes are recorded in CHANGELOG.md

API conventions

No secrets or tokens committed to source

Security

Migrations are reversible (down() implemented)

Data quality

New endpoints have OpenAPI specs

API conventions

Pydantic models declare types for every field

Type safety

React components render under StrictMode

Frontend hygiene

Logs include request_id and tenant_id

Observability

No new try/except without an explicit reason

Error handling

PII columns are tagged in the data catalog

Compliance

Dependency bumps are pinned to exact versions

Supply chain

Pydantic models declare types for every field

Type safety

React components render under StrictMode

Frontend hygiene

Logs include request_id and tenant_id

Observability

No new try/except without an explicit reason

Error handling

PII columns are tagged in the data catalog

Compliance

Dependency bumps are pinned to exact versions

Supply chain

Run viewer

What you actually get back, on every PR.

Per-statement results, grouped, ranked by issues, with line-numbered evidence. Try the failing row — it expands into the source.

Data quality

2 checks

101

All new tables have a Great Expectations test suite

94%

Fail

Migration 0042 introduces table `events_session` with no matching expectation suite under tests/great_expectations/.

migrations/0042_events_session.sql:1-12

1-- Add events_session table for per-session aggregation
2CREATE TABLE events_session (
3    session_id   UUID PRIMARY KEY,
4    user_id      UUID NOT NULL REFERENCES users(id),
5    started_at   TIMESTAMPTZ NOT NULL,
6    ended_at     TIMESTAMPTZ,
7    event_count  INTEGER NOT NULL DEFAULT 0,
8    properties   JSONB
9);
10 
11CREATE INDEX events_session_user_id_idx ON events_session (user_id);
12CREATE INDEX events_session_started_at_idx ON events_session (started_at);
13

Migrations are reversible (down() implemented)
96%
Pass

API conventions

2 checks

011

Every new public function has a docstring
99%
Pass
Public API changes are recorded in CHANGELOG.md
71%
Needs review
PublicSearchClient.search() signature changed; CHANGELOG.md was not updated. May be intentional if pre-1.0.
src/public/search.ts:18-24

Security

1 check

001

No secrets, tokens, or .env values in source
97%
Pass

Extensibility

Tools are data, not code.

Wire up MCP servers, custom skills, or the built-in repo toolset from the admin UI. Pick a review depth per repo; CrossCheck owns the model + cost story. No deploy required.

Three review depths — Fast / Standard / Deep — pick the cost-vs-thoroughness tradeoff per repo.

Built-in tools for navigating your repo — reading files, searching, running commands in an isolated checkout.

Plug in MCP servers for org-specific knowledge — Linear, Slack, your own.

Skills let you bundle instructions + tools and reuse them across check groups.

FAQ

Things people ask before they sign up.

In the CrossCheck-managed cloud. Each PR spawns a fresh, isolated review that checks out your code, evaluates the statements, and exits — nothing persists between runs. Install the CrossCheck GitHub App once; there's no CI workflow to maintain.

Doors to manual and cross check

Cut the “is this up to standard?” question out of code review.

Encode your team’s standards as statements. Ship the action. Get pass/fail with evidence on every PR.

Start free trial - no card Sign in

14-day Pro trial. No credit card required.