[SanderDeclerck]

No description provided.

[aikido-pr-checks]

🟡 Medium - Over-limit gzip metadata bypasses minimum-age filtering

A registry can return a gzip metadata response whose decoded size exceeds the new 256 MiB limit. The decoder rejects it, but the response handler catches that error and forwards the original compressed bytes without invoking the metadata interceptor, so npm or pip receives unfiltered version/file metadata and can select releases that the minimum-age rewrite was meant to remove.

Show fix

Treat an over-limit decode as a failed upstream response and do not forward it to the package manager, or use a bounded streaming rewrite path that can safely enforce the limit while still applying the metadata checks. Add an integration test proving that an over-limit gzip response cannot reach the client as an unmodified metadata response.

_{More info - Reply on this comment to give feedback or ignore the issue.}

[aikido-pr-checks]

Partially fixed in commit 7f82be1.

This issue remains because packages/safe-chain/src/registryProxy/mitmRequestHandler.js:270-278 catches the MAX_DECODED_BODY_BYTES rejection from decodeBody and sends status/headers plus originalBuffer, allowing the over-limit compressed metadata to reach the client unmodified.

[aikido-pr-checks]

🔵 Low - Accept-Encoding wildcard with identity refusal is rewritten to an unacceptable encoding

A client that sends Accept-Encoding: *;q=1, identity;q=0 can accept gzip, but supportedAcceptEncoding drops the wildcard and falls back to identity. The proxy therefore asks the registry for identity and forwards an identity metadata response, violating the client's explicit refusal and causing clients that enforce their encoding negotiation to abort the metadata request or installation.

Show fix

Suggested change

	export function supportedAcceptEncoding(acceptEncoding) {
	const offered = String(acceptEncoding ?? "")
	.split(",")
	.map((token) => token.trim())
	.filter(Boolean);
	const supported = offered.filter((token) =>
	Object.hasOwn(decoders, token.split(";")[0].trim().toLowerCase())
	);
	return supported.length > 0 ? supported.join(", ") : "identity";
	}
	export function supportedAcceptEncoding(acceptEncoding) {
	const entries = String(acceptEncoding ?? "").split(",").map((token) => {
	const [name, ...params] = token.trim().split(";");
	const q = params.find((param) => param.trim().toLowerCase().startsWith("q="));
	return { token, name: name.trim().toLowerCase(), q: q ? Number(q.split("=")[1]) : 1 };
	}).filter(({ name }) => name);
	const supported = entries.filter(({ name, q }) => Object.hasOwn(decoders, name) && q > 0);
	if (supported.length) return supported.map(({ token }) => token).join(", ");
	const wildcard = entries.find(({ name, q }) => name === "*" && q > 0);
	const blocked = new Set(entries.filter(({ q }) => q === 0).map(({ name }) => name));
	const wildcardEncoding = wildcard && Object.keys(decoders).find((name) => !blocked.has(name));
	if (wildcardEncoding) return `${wildcardEncoding}${wildcard.q === 1 ? "" : `;q=${wildcard.q}`}`;
	const identity = entries.find(({ name }) => name === "identity");
	if (!identity || identity.q > 0) return "identity";
	throw new Error("No acceptable content-encoding supported");
	}

_{More info - Reply on this comment to give feedback or ignore the issue.}