Four Platforms, Four Meanings of “Access”
Agent Reach’s route policy becomes most legible when the same word—“access”—is tested against different platforms. This chapter compares Bilibili, Reddit, XiaoHongShu, and Twitter by the criteria that actually determine a route: the operation requested, the environment in which it runs, the authentication state it may use, the evidence available, and the feature coverage a fallback preserves. The aim is not to crown one backend, but to show why a platform-specific route must remain conditional and time-sensitive.
Bilibili, Reddit, and XiaoHongShu: Same Word, Different Route
The first comparison is between three platforms whose access problems look similar from a distance. Each can involve anti-bot defenses, changing upstream tools, and a choice between direct commands and browser-backed routes. Yet their current routes are different because the failure mechanism and the available evidence are different.
Bilibili demonstrates a platform-specific retirement. The repository records that requests from yt-dlp were receiving HTTP 412 under Bilibili’s risk controls. That observation changed the Bilibili route: yt-dlp is no longer used there, while bili-cli remains the preferred path for search, popular content, and video details without login. OpenCLI can cover subtitles through a browser session. The important point is not that one tool is “bad.” yt-dlp remains the documented YouTube backend. Its suitability changed because the platform and operation changed.
This is a useful test for the meaning of a fallback. If a user asks Bilibili for search results, bili-cli may preserve the requested capability without a login. If the user asks for subtitles, the route may move to OpenCLI and inherit the conditions of a desktop browser bridge. A search API fallback can preserve discovery while failing to provide video details. A metadata route can confirm that a video exists while failing to return captions. The platform name stays constant; the capability does not.
The Bilibili case also shows why a route policy needs negative knowledge. It is not enough to record that a tool once worked. The channel must know that a tempting generic path should not be reused for this platform under the observed conditions. That knowledge is time-sensitive: the repository describes a live observation in a particular snapshot, not a permanent law about every future version of Bilibili or yt-dlp. The route is maintainable because the restriction is attached to the Bilibili channel rather than generalized to every video source.
Reddit presents a different constraint: there is no anonymous route to fall back to in the current design. The channel states that anonymous JSON endpoints are blocked and that the official API requires manual approval. A user cannot solve that problem by installing one more public-reader package. Both current routes depend on a logged-in session: OpenCLI can reuse a desktop browser session, while rdt-cli serves server or legacy use after an explicit session or cookie arrangement. The route choice is therefore not “free API versus paid API.” It is “which logged-in session path fits the environment and the user’s authority?”