can1357/oh-my-pi · error

Path traversal (..) is not allowed in omp:// URLs

Error message

Path traversal (..) is not allowed in omp:// URLs

What it means

As part of omp:// path sanitization, after normalizing the filename the handler rejects any result equal to '..' or containing '../' segments, blocking directory traversal out of the bundled docs namespace.

Source

Thrown at packages/coding-agent/src/internal-urls/omp-protocol.ts:65

		const content = `# Documentation\n\n${filenames.length} files available:\n\n${listing}\n`;

		return {
			url: url.href,
			content,
			contentType: "text/markdown",
			size: Buffer.byteLength(content, "utf-8"),
		};
	}

	async #readDoc(filename: string, url: InternalUrl): Promise<InternalResource> {
		// Validate: no traversal, no absolute paths
		if (path.isAbsolute(filename)) {
			throw new Error("Absolute paths are not allowed in omp:// URLs");
		}

		const normalized = path.posix.normalize(filename.replaceAll("\\", "/"));
		if (normalized === ".." || normalized.startsWith("../") || normalized.includes("/../")) {
			throw new Error("Path traversal (..) is not allowed in omp:// URLs");
		}

		const docPath =
			normalized === "docs" ? "" : normalized.startsWith("docs/") ? normalized.slice("docs/".length) : normalized;
		if (!docPath) {
			return this.#listDocs(url);
		}

		const content = await getEmbeddedDoc(docPath);
		if (content === undefined) {
			const lookup = docPath.replace(/\.md$/, "");
			const suggestions = getDocFilenames()
				.filter(f => f.includes(lookup) || lookup.includes(f.replace(/\.md$/, "")))
				.slice(0, 5);
			const suffix =
				suggestions.length > 0
					? `\nDid you mean: ${suggestions.join(", ")}`
					: "\nUse omp:// to list available files.";

View on GitHub (pinned to 9690622007)

Solutions

  1. Remove '..' segments and reference the doc file by its plain name within docs/.
  2. Sanitize/normalize user-supplied names before embedding them into omp:// URLs.
  3. Use omp:// with no path to list valid doc filenames.

Example fix

// before
router.resolve(`omp://${userInput}`);
// after
const safe = path.posix.normalize(userInput.replaceAll('\\', '/'));
if (safe === '..' || safe.startsWith('../') || safe.includes('/../')) throw new Error('invalid doc name');
router.resolve(`omp://${safe}`);
Defensive patterns

Strategy: validation

Validate before calling

const normalized = path.posix.normalize(input.replaceAll('\\', '/'));
if (normalized === '..' || normalized.startsWith('../') || normalized.includes('/../')) throw new Error('traversal not allowed in omp:// URL');

Type guard

const isTraversalSafe = (s: string): boolean => {
  const n = path.posix.normalize(s.replaceAll('\\', '/'));
  return n !== '..' && !n.startsWith('../') && !n.includes('/../');
};

Try / catch

try {
  return await router.resolve(url);
} catch (err) {
  if (err instanceof Error && err.message.includes('Path traversal')) throw new Error('omp:// doc names cannot contain ..');
  throw err;
}

Prevention

When it happens

Trigger: Resolving omp://../secrets or omp://docs/../../etc/passwd — the normalized path escapes the docs root via '..' segments.

Common situations: Programmatically joining user input into an omp URL; untrusted config files embedding traversal paths; URL-encoding tricks that decode to '..'.

Related errors


AI-assisted analysis of can1357/oh-my-pi@9690622007 (2026-08-31). Data as JSON: /api/errors/6567c2c85ddfa98b. Report an issue: GitHub.