Skip to content

d3v07/lintellect

Repository files navigation

Lintellect

Serverless pull request review pipeline that turns a GitHub webhook into parallel review packets, evidence-gated findings, and an operator dashboard.

Lintellect is a TypeScript monorepo for experimenting with review automation architecture: webhook ingestion, diff/context workers, provider adapters, schema validation, prompt runners, an evidence gate, and admin/dashboard surfaces. The repo is structured so the review pipeline can be tested locally while the deployment target remains AWS Lambda, Step Functions, S3, DynamoDB, and API Gateway.

Contents

At A Glance

Area Details
Product Event-driven PR review pipeline
Users Engineering teams that want structured review findings with evidence
Core value Split review work into traceable packets and filter output through a quality gate before comments are posted
Runtime target AWS Lambda + Step Functions
Local code TypeScript workspaces with Vitest coverage
UI surface Admin/dashboard packages for configuration and review visibility

Screenshot Gallery

Image Caption
Lintellect setup screen Initial setup state for configuring provider and review pipeline settings before packet execution.
Lintellect activity dashboard Authenticated dashboard with review counts, monthly usage, pipeline stages, and recent review activity.
Lintellect pull request list Pull request queue with connected repository selector, review filters, and reviewed PR indicators.
Lintellect PR review detail PR detail view with changed files, evidence-gated findings, severity labels, and review actions.
Lintellect repository settings Repository settings screen for connecting repositories and managing review coverage.

Review Lifecycle

flowchart LR
  A["GitHub pull request event"] --> B["Webhook handler validates signature"]
  B --> C["Step Functions execution"]
  C --> D["Diff worker extracts changed hunks"]
  C --> E["Context worker gathers surrounding code"]
  D --> F["Review packet builder"]
  E --> F
  F --> G["Prompt/provider runner"]
  G --> H["Evidence gate"]
  H --> I["Comment formatter"]
  I --> J["GitHub PR comment"]
Loading

Architecture

flowchart TD
  GH["GitHub Webhook"] --> APIGW["API Gateway"]
  APIGW --> WH["webhook Lambda"]
  WH --> SFN["Step Functions state machine"]
  SFN --> DIFF["diff-worker Lambda"]
  SFN --> CTX["context-worker Lambda"]
  DIFF --> S3["S3 review artifacts"]
  CTX --> S3
  S3 --> REVIEW["review-worker Lambda"]
  REVIEW --> MERGE["merge-results Lambda"]
  MERGE --> GATE["evidence-gate Lambda"]
  GATE --> POST["comment-poster Lambda"]
  POST --> PR["Pull request discussion"]
  GATE --> DDB["DynamoDB review history"]
  DASH["Dashboard API"] --> DDB
  ADMIN["Admin UI"] --> DASH
Loading

Package Map

Path Responsibility
packages/core Review packet building, schema validation, prompt runner, evidence gate tests
packages/providers Provider abstraction and provider tests
packages/api Dashboard API entrypoint
packages/cli Local command surface
packages/admin Admin UI
packages/dashboard Dashboard UI
schemas JSON schemas for packets, comments, job status, and provider config
docs Pipeline, architecture, prompting, and testing notes

Schemas

The pipeline is schema-first. Review packets, review output, comments, job status, and provider config are stored in schemas/ so worker boundaries can validate structured data before a result reaches the evidence gate.

Tech Stack

Layer Technology
Language TypeScript
Infrastructure AWS CDK target, Lambda, Step Functions, API Gateway
Storage S3 artifacts, DynamoDB history
Frontend React/Vite packages
Testing Vitest
Contracts JSON Schema

Run Locally

npm install
npm test

Build all workspaces:

npm run build

Run the dashboard/admin package from its package directory when working on UI surfaces.

Verification

Local build results:

Command Result
npm run build in packages/admin Passed in the previous docs/screenshot pass
npm run build in packages/dashboard Passed in the current README/build verification slice

The checked dashboard build now passes. Core tests and package-specific tests should be run before publishing.

Deployment Notes

Production deployment expects:

GITHUB_WEBHOOK_SECRET=
GITHUB_TOKEN=
AWS_REGION=us-east-1
DYNAMO_TABLE_NAME=
S3_BUCKET_NAME=

Keep webhook secrets and provider credentials in environment variables or a secret manager.

Status

The architecture and core contracts are useful and portfolio-worthy. The checked dashboard build now passes; before publishing, run core/package tests and verify deployment configuration with real provider credentials kept outside git.

License

MIT. See LICENSE.

About

Automated code review agent — AWS CDK Step Functions pipeline, multi-provider LLM integration, evidence validation

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors