Project X-Ray

A metadata-only map that shows AI agents where to start in unfamiliar codebases—without reading or uploading source.

Comment

Inspiration

AI agents can lose time and context by blindly crawling an unfamiliar repository. Project X-Ray gives an agent an orientation layer before it starts reading or editing code.

What it does

Project X-Ray scans browser-visible file and folder metadata—names, paths, types, sizes, and hierarchy—not file contents. It creates a bounded, ordered Start Here route; Avoid First context; direct proof paths separated from supporting validation metadata; a navigable Project Map designed to stay bounded for large repositories; and a five-part AI Agent Brief. The result can be copied, downloaded as Markdown, or exported as JSON.

Privacy

Project X-Ray’s scanner is metadata-only. It does not read or upload project file contents. The public page separately checks the same-origin download-account session, but the repository scan itself remains browser-local.

What we built during Build Week

Project X-Ray existed before Build Week: the browser-local metadata scanner, AI Agent Brief concept, and export surfaces were already present. The defensible pre-Build Week baseline is commit 7dfcce9, before the July 13, 2026 9:00 AM Pacific cutoff.

During Build Week, we added meaningful product and quality work:

  • hardened native folder selection;
  • improved report quality and Start Here/proof semantics;
  • introduced and refined a collapsible Project Map with bounded incremental rendering for large repositories;
  • expanded browser-quality coverage for privacy, mobile behavior, exports, route quality, and large synthetic projects; and
  • added security-oriented deployment bridge and Linux-toolchain work.

Evidence, in commit date order (EDT): b2c9b74 (Jul 13, PR #239); 06d087e (Jul 15, PR #242); 1a01e06 (Jul 16, PR #243); ed117a6 (Jul 18, PR #250); 606320c and 293538a (Jul 19, PRs #251–252); d0bb726 and 631e25f (Jul 20, PRs #253–254).

The deployment tooling is merged source work; it is not a claim that the hardened deployment candidate has completed its remaining production gate.

How we built it

The public tool is a browser-based HTML/CSS/JavaScript experience. Browser-quality coverage uses Node and Playwright; deployment-security coverage uses Python fixture tests.

Codex with GPT-5.6 was used in the Build Week development workflow for repository inspection, implementation and review iterations, and quality/deployment work. That workflow produced the folder-picker, report-quality, Project Map, test, and deployment-safety additions above. Project X-Ray does not call GPT-5.6 at runtime. Its scanner works from local metadata and does not read or upload project file contents.

What’s next

We will keep strengthening reproducible judging documentation, privacy-preserving large-repository behavior, and deployment verification while preserving the metadata-only boundary.

Built With

Share this project:

Updates