AI-assisted scope control for software projects.
Turn ambiguous client requests into structured scope-risk reviews, impact estimates, and clearer change-order conversations.
ScopeGuard is a project-scope auditing prototype for software teams, freelancers, and delivery leads. It compares a new client request against a project's Statement of Work (SOW) and produces a structured assessment that can help a team decide whether a request appears in scope, ambiguous, or likely to require a change order.
Project status: AI-assisted decision support, not legal advice or an automated contract interpretation system. Review the source SOW, assumptions, estimates, and generated client communications before using them in a real engagement.
- SOW-aware review — evaluates client requests against project-specific deliverables, exclusions, and change-order clauses.
- Scope verdicts — returns
IN_SCOPE,AMBIGUOUS_TRAP, orOUT_OF_SCOPE_CREEP. - Impact breakdown — provides a risk score, estimated additional hours, revenue-at-risk estimate, and line items.
- Change-order draft — generates a client-facing email draft to help clarify and document requested changes.
- Scenario presets — try representative examples, including an administrative portal request, payment integration work, and a small visual adjustment.
- Model telemetry — surfaces model name, latency, token counts, estimated API cost, and trace ID when returned by the application.
- Fallback analysis — includes deterministic handling for selected request patterns when the model-backed path is unavailable or unsuitable.
- SOW retrieval — supports project SOW records and a local fallback mode.
flowchart TD
A[Client request] --> B[ScopeGuard dashboard]
B --> C[Select project SOW]
C --> D[SOW retrieval]
D --> E[AI-assisted analysis]
D --> F[Deterministic fallback rules]
E --> G[Structured scope assessment]
F --> G
G --> H[Verdict and risk score]
G --> I[Effort and cost estimate]
G --> J[Change-order email draft]
G --> K[Telemetry and trace ID]
The analysis is intended to support human review. The estimated effort and cost are illustrative outputs based on the configured project data and analysis logic; they are not guarantees or contractual quotes.
| Layer | Technology |
|---|---|
| Web application | Next.js App Router, React, TypeScript |
| AI workflow | Mastra, Vercel AI SDK, OpenAI-compatible provider |
| Validation | Zod |
| SOW storage | MongoDB with local fallback support |
| Error monitoring | Sentry for Next.js |
- Node.js compatible with the Next.js version in
package.json - npm (or another compatible Node package manager)
- An AI provider key for the model-backed analysis path
- MongoDB configuration if you want to use remote SOW storage
git clone https://github.com/kishorekrrish3/ScopeGuard.git
cd ScopeGuard
npm installCreate a local .env.local file using the variable names expected by the project configuration. Keep API keys and database credentials out of Git.
npm run devOpen http://localhost:3000.
npm run dev
npm run build
npm run start
npm run lint- Choose a project and its SOW.
- Enter the new client request or select a scenario preset.
- Run the scope audit.
- Review the verdict, cited clause, reasoning, effort estimate, and line items.
- Edit and approve any change-order communication before sending it.
src/
├── app/ # Dashboard and API routes
├── lib/ # SOW contracts and data access
└── mastra/ # Agent and tool definitions
- Treat generated verdicts, risk scores, hours, and dollar estimates as suggestions to review—not authoritative determinations.
- Confirm that the correct SOW version, currency, rates, and clauses are configured for each project.
- Do not submit confidential client material unless you are authorized to process it with the configured provider and storage system.
- Keep credentials in environment variables and use least-privilege access for any connected database or AI provider.
- ScopeGuard is not a substitute for legal, commercial, or project-management judgment.
- Add a richer test suite for ambiguous and adversarial requests.
- Improve traceable links from each finding to the exact SOW clause.
- Add exportable review reports and an approval history.
- Expand configuration and documentation for provider and storage modes.
Issues and pull requests are welcome. Please include reproducible steps for bugs and clearly label assumptions in proposed scope-analysis rules.
Built as an exploration of AI-assisted project governance and scope management.