Runway: Pre-Commit Intelligence for Engineers

Catch risky code changes before you commit.

Runway is a staff-engineer-style code reviewer that runs in your IDE — not in a PR, not in the cloud. It combines deterministic signal detection with AI analysis to surface architectural decisions, security risks, API contract changes, and repo-pattern drift before they become problems.

Bring your own API key. Your code never leaves your machine except to the AI provider you configure.

Why Runway

Runway runs before you commit, when fixes are cheapest. It focuses on architectural risk, decisions, staged/unstaged state, and repo-pattern drift — not style nitpicking.

Opinionated, terse, and evidence-backed. Every finding includes a file path, function name, or diff reference — no guesses.

What it detects

⚡ Deterministic Signals (no AI required)

Fires instantly from your working tree — no API call needed:

• Reversals — commit messages suggesting rollback, deprecation, or removal
• Architectural conflicts — e.g. GraphQL vs REST, monolith vs microservice in the same change
• Hotspots — files with high commit churn that carry regression risk
• AI-generated code — detects AI-assisted code via co-authored-by trailers and comment density
• Missing tests — large changes with no test file modifications
• Coupled files — files that historically change together but weren't included
• Security patterns — hardcoded secrets, eval(), XSS risks, SQL injection patterns
• Large changes — changes spanning too many files or lines to review safely

🤖 AI Analysis (Claude or GPT)

Runs on top of deterministic signals for deeper insight:

• Architectural decisions — surfaces design decisions from your commits (new dependency, API contract change, schema migration, auth model change)
• Additional risks — catches what deterministic checks miss: correctness issues, migration hazards, product risk, API contract drift
• Verdict — one-sentence overall assessment, evidence-backed

🔍 Review My Changes

Full AI review of staged + unstaged changes before commit:

• Verdict: Ready to Commit or Needs Work
• Risk level: HIGH / MEDIUM / LOW with specific reasons
• Blocking issues — bugs, security risks, data loss, broken contracts
• Non-blocking suggestions — maintainability and design improvements
• Evidence for every finding: file path, function name, or diff line

🗺 Codebase Intelligence

Deep analysis of your repo using actual file content:

• Architecture patterns, tech stack, risk areas, code health
• Saved locally per repo — enriches every future scan with your codebase context
• No re-analysis needed on every run

🔥 Editor Decorations

After a scan, open any changed file to see churn indicators in the gutter:

• 🔥 High-churn file (≥7 commits in 30 days)
• ⚡ Active file (≥4 commits in 30 days)

Getting Started

1. Install the extension
2. Open a Git repository in VS Code
3. Click the ⚡ Runway icon in the activity bar
4. Go to Settings and add your API key:
   • Anthropic Claude — recommended
   • OpenAI GPT
5. Run Codebase Intelligence once to build local context
6. Run Signals or Review My Changes before every commit

Privacy

All analysis runs locally. Your code is sent only to the AI provider you configure (Anthropic or OpenAI) using your own API key. No data is sent to Collabsuite servers. Codebase context is stored as a local JSON file on your machine.

Git Hooks (optional)

Install pre-commit or pre-push hooks from Settings to trigger analysis automatically on every commit or push.

Requirements

• VS Code 1.85.0 or later
• Git installed and available in PATH
• An Anthropic or OpenAI API key

Commands

Support

For issues and feedback: support@collabsuite.com

© 2025 Collabsuite. All rights reserved.