CoStrict Documentation

The architecture is composed of several specialized components:

• Next.js Web Application: The management interface for creating runs, monitoring progress via Server-Sent Events (SSE), and viewing the results dashboard.
• Controller Container: An orchestrator (based on evals-runner) that manages an in-memory task queue using p-queue, prepares Git workspaces, and coordinates runner containers.
• Runner Containers: Isolated environments that execute single tasks. They include a fresh VS Code instance, pre-installed language runtimes (Go, Java, JavaScript, Python, Rust), and the Roo Code extension.
• Redis: Used for pub/sub messaging, runner registration tracking, and heartbeat monitoring (Note: it is not used for task queuing).
• PostgreSQL: The primary database for storing run configurations, task definitions, metrics, and results.

CoStrict is an open-source AI-assisted programming tool designed for enterprise-level development. It supports private deployment and focuses on standardized, secure, and high-quality AI development workflows.

Core Capabilities:

• Strict Mode: Standardized AI code generation including requirement analysis, architecture design, task planning, and test generation.
• Code Review: Repository-wide RAG-based analysis using multi-expert model cross-verification.
• Code Completion: Context-aware code generation.
• MCP Integration: Supports connecting to APIs, databases, and custom tools via Model Context Protocol.
• Multimodal Support: Supports image uploads and visual context.
• Extensible Skills: A system for specialized task workflows.

CoStrict is an enterprise-grade AI coding assistant providing the following core features:

• Strict Mode: Standardized AI code generation involving requirements analysis, architecture design, task planning, and test generation.
• Code Review: Repository-wide RAG-based analysis using multi-expert model verification.
• Code Completion: Context-aware generation.
• Vibe Code: Rapid development via natural language dialogue.
• MCP Integration: Connectivity for APIs, databases, and custom tools via Model Context Protocol.
• Multi-modal Support: Ability to process image context and visual inputs.
• Skills Support: An extensible system for specialized task workflows.

The Roo Code Cloud Terms of Service govern the use of Roo Code software and services. By clicking "Sign Up", "I agree", or otherwise indicating assent, or by accessing/using any Software, you enter into a legally binding agreement with Roo Code, Inc.

Key aspects include:

• Agreement Scope: Includes the Data Processing Addendum, Privacy Policy, and any applicable Entity Agreement.
• Limited License: Users are granted a limited, revocable, non-exclusive, non-sublicensable, and non-transferable license to use the Software for purposes specified in an Order Form.
• No Reliance: Users should not rely on promises of future functionality or features.
• Data Privacy: Users in Europe (EEA, UK, Switzerland) are subject to specific privacy acknowledgments. For users outside Europe, the agreement includes the Roo Code Data Protection Notice.

Roo Code facilitates organic adoption by providing internal influencers with repeatable and reviewable outcomes. Unlike theoretical best practices, Roo Code allows influencers to demonstrate real, reproducible workflows in demo forums using explicit artifacts:

• Diffs: Showing exact code changes.
• Test Output: Demonstrating successful validation.
• Execution Trails: Providing a clear record of what was attempted, including failure modes.

This makes it easier for influencers to share exactly what worked (and what didn't) with their peers.

To introduce a completely new language to the Roo Code evals system, you must update multiple components:

1. Update Language Configuration: Add the language name to the exerciseLanguages array in packages/evals/src/exercises/index.ts.
2. Create Language Prompt: Create prompts/{language}.md in the evals repository. This prompt should instruct the AI to complete the exercise and specify the {test-command} used to verify correctness.
3. Update Docker Configuration: Modify packages/evals/Dockerfile.runner to include the necessary runtime and dependencies for the new language.
4. Update Test Runner Integration: If the language uses a non-standard testing framework, update packages/evals/src/cli/runUnitTest.ts to handle it.
5. Seed Exercises: Create 2-3 initial exercises using the standard exercise structure to validate the setup.

While prototype-first workflows increase velocity, they introduce specific failure modes that require mitigation:

• Scope Creep: Without an initial constraint document, prototypes can drift.
  • Mitigation: Use a lightweight one-page brief that states the problem, success criteria, and hard boundaries to anchor scope.
• Lost Rationale: Losing the 'why' behind decisions can damage institutional memory.
  • Mitigation: Use lightweight decision records such as ADRs (Architecture Decision Records), RFC-style docs, structured commit messages, or task logs generated by tools like Roo Code.
• Stakeholder Whiplash: Stakeholders may lack trust without a polished plan.
  • Mitigation: Align stakeholders around working software they can touch and interact with, rather than abstract documents.

The Body is an optional detailed description of the commit. It should be written in the present tense and must explain the motivation for the change and how it differs from previous versions.

Safari 15.4+ enforces stricter CORS policies for localStorage access,
causing intermittent auth failures. Added retry logic for token refresh.