[Toyenzo]

Description

This covers items 1 and 2 of #3024. DirectUrlDocumentLoader and CalmHubDocumentLoader now share one request-path allowlist and one directory-traversal guard, in document-loader/request-guards.ts. It follows the response-origin.ts precedent. A future tightening now applies to both loaders.

While writing the shared guard I found that the old check, documentId.includes('/..'), only caught a literal ... The WHATWG URL parser also resolves these forms to ..:

• %2e%2e, %2E%2E, .%2e, %2e.
• \..\ in http(s) URLs, where the parser treats \ as /
• dots split by a tab or newline, which the parser strips

By the time either loader read url.pathname, the parser had already resolved the segment away. So https://calm.finos.org/calm/%2e%2e/secret.json passed the guard and was requested as /secret.json. The same happened with calm:/schemas/2025-03/%2e%2e/%2e%2e/admin/... against CalmHub. The shared guard applies the same rewrites as the parser before it looks for /... Plain URLs behave as before.

Notes:

• The two allowlists were not actually different. The CalmHub pattern's optional (\.json)? suffix was already matched by the character class, so one pattern keeps today's behaviour for both loaders.
• The CalmHub error messages now read "rejected a path ...", the same wording as the direct-URL loader. Nothing in the repo matched the old wording.
• Items 3 (DNS resolution and pinning), 4 (axios debug hooks) and 5 (double loadSchemas()) are left for follow-ups. Item 3 needs the Node/browser design agreed first, as the comment on the issue says. I used "Refs" so the issue stays open.

Refs #3024

AI-assisted: drafted with Claude Code; reviewed and tested by me.

Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)
• ✨ New feature (non-breaking change which adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📚 Documentation update
• 🎨 Code style/formatting changes
• ♻️ Refactoring (no functional changes)
• ⚡ Performance improvements
• ✅ Test additions or updates
• 🔧 Chore (maintenance, dependencies, CI, etc.)

Affected Components

• CLI (cli/)
• Schema (calm/)
• CALM AI (calm-ai/)
• CALM Hub (calm-hub/)
• CALM Hub UI (calm-hub-ui/)
• CALM Lab (calm-lab/)
• CALM Models (calm-models/)
• CALM Server (calm-server/)
• CALM Studio (calm-studio/)
• CALM Widgets (calm-widgets/)
• CALMGuard (calm-guard/)
• Documentation (docs/)
• Shared (shared/)
• VS Code Extension (calm-plugins/vscode/)
• Dependencies
• CI/CD

Commit Message Format ✅

Testing

• I have tested my changes locally

• I have added/updated unit tests

• All existing tests pass

• 29 new tests: 19 unit tests for the shared guards, plus 5 per loader for the parser-resolved traversal forms. The loader tests also check that no HTTP request is sent.

• All 10 new loader tests fail against the old inline checks. The old code sent the request and failed later with "Failed to load document ...".

• npm run build:cli passes. npm test --workspace shared passes 1573/1573, with coverage thresholds met and the browser-entry check OK.

• CLI: 982 unit tests pass. cli.e2e passes 60/60 when run with --hookTimeout=300000. With the default 30s hook timeout it times out locally while packing and installing the CLI, which is unrelated to this change.

• npm run lint: 0 errors. The only warnings are existing ones in untouched packages.

Checklist

• My commits follow the conventional commit format
• I have updated documentation if necessary (shared/AGENTS.md, Document Loader section)
• I have added tests for my changes (if applicable)
• My changes follow the project's coding standards

🤖 Generated with Claude Code