fix(security): sandbox repo verification in Docker to prevent RCE (#211)

[arpita-1111]

Linked issue

Closes #211

What this PR does

Prevents Remote Code Execution (RCE) during repository verification by running all untrusted build/test/lint commands inside an isolated, disposable Docker container instead of directly on the host.

Previously, verify_repo() executed npm install, npm ci, npm run <script>, and pytest directly on the host machine using scripts defined in the uploaded repository's package.json. Since these repositories are untrusted, a malicious package.json script could execute arbitrary commands with full access to the host system, environment variables, and network.

Type of change

• Bug fix
• New feature
• ML model / training pipeline
• Refactor (no behaviour change)
• Documentation
• Tests only
• Security fix

Stack affected

• Backend
• Frontend
• Both

---

Changes

Backend

• Added run_cmd_sandboxed() in backend/app/utils/exec.py, which runs a command inside an ephemeral Docker container instead of on the host.
• Container is configured with:
  • --rm (self-destructs after running)
  • --user 1000:1000 (non-root)
  • --memory 512m and --cpus 1 (resource limits)
  • --pids-limit 256 (prevents fork bombs)
  • --network none by default (no network access)
  • --cap-drop ALL and --security-opt no-new-privileges (minimal Linux capabilities)
  • Repository mounted read/write only inside /workspace
• Updated backend/app/sandbox/verify.py to use run_cmd_sandboxed() instead of run_cmd() for all commands that execute code from the uploaded repository (npm ci, npm install, npm run <script>, pytest).
• Added an explicit check for Docker availability on the host; verification fails gracefully with a clear reason if Docker isn't installed, instead of silently running unsandboxed.
• Left the final fallback check (python -c "print('verify: no tests detected')") unsandboxed since it only prints a fixed string and never touches repository content.

Proposed solution checklist (from issue)

• Repository verification no longer executes untrusted scripts on the host system.
• Verification runs inside an isolated container (Docker).
• Sandbox enforces resource limits (CPU, memory).
• Verification runs as a non-privileged user.
• Network access is disabled by default.
• Temporary verification environments are cleaned up after completion (--rm).
• Existing verification functionality continues to work for legitimate repositories (pending live scan test).

---

Testing

How did you test this?

• Installed Docker Desktop + WSL2 locally.
• Ran docker run hello-world to confirm the Docker engine works.
• Pulled the node:20-alpine and python:3.11-slim base images used by the sandbox.
• Reviewed the code path manually to confirm run_cmd is no longer called for any repository-controlled command.

Checklist

• Tested locally end-to-end (upload ZIP or GitHub URL → scan → findings returned correctly)
• New ML model falls back gracefully when model file is absent
• No new console.error or unhandled Python exceptions introduced
• Added or updated tests where applicable
• requirements.txt / package.json updated if new dependencies added
• New model files (.pkl, .pt, etc.) are gitignored, not committed

---

Anything reviewers should focus on

Please confirm the resource limits (memory/CPU/pids) are reasonable for real-world verification workloads, and whether the sandbox should be given network access in specific cases (e.g., npm registry access during npm ci/npm install, which currently would fail with --network none). Happy to adjust based on maintainer guidance.

Screenshots (if UI changed)

N/A (backend-only change)

[github-actions]

🎉 Thank you @arpita-1111 for submitting a Pull Request!

We're excited to review your contribution.

✅ Before Review

• Ensure all CI checks pass.
• Complete the PR template.
• Link the related issue.

⚡ Want faster reviews and contributor support?

Join our Discord community:

🔗 https://discord.gg/FcXuyw2Rs

Maintainers and mentors are active there and can help resolve blockers quickly.

Happy Contributing! 🚀