Clarity for AI buildersSponsored by AI Routing
adoption

Chris Greening shares a Claude brief for building a watch-history manager, not its source code

Chris Greening shares a Claude brief for building a watch-history manager, not its source code

Chris Greening describes using Claude to build a watch-history app, sharing a rebuild brief rather than releasing its code.

In an Atomic14 post, Greening explains how frustration with Prime Video’s watch-history controls led him to ask Claude first to operate the website, then to write a script, and finally to build an app. His post offers implementation notes for managing Prime Video and Netflix history, rather than a downloadable application or an open-source release.

The problem he wanted to solve was straightforward: when rewatching a series, Prime resumed episodes near the closing credits instead of starting them again. Greening describes the web watch-history interface as an infinitely scrolling list without filters or bulk actions. His app was intended to make that history easier to browse and delete in batches.

How the proposed app works

Greening’s brief calls for an embedded web view to handle authentication. When signed out, the app should show the service’s login page and close the login dialog after returning to the history page. He explicitly instructs the app not to touch the password. Requests then run as JavaScript fetch calls inside that web view, using the existing session cookies.

For Prime Video, he specifies a configurable regional base URL and the getWatchHistorySettingsPage endpoint. According to the post, responses contain day groups, title records and a nextToken for pagination. Individual titles and episode children carry REMOVE actions with the identifiers and tokens needed for deletion. The proposed app submits those values to activityHistoryTitleUpdate.

This makes the tool a layer over the account’s existing web session, rather than a separate watch-history service. It also means the app depends on the structure and behavior of the service’s website: a custom interface does not remove the underlying integration dependency.

Keeping local history consistent

The brief stores history as JSON in the app’s data folder, keyed by day and the title’s gti identifier. Incremental synchronization reads newest-first until reaching the newest day already stored, then replaces the overlapping days. A separate full re-sync provides a way to rebuild the local record.

Greening also describes normalizing season names, including removing season suffixes and tags, so separate seasons appear under one show. This addresses a distinction between the service’s records and the interface he wants: Prime supplies season-level entries, while the app groups them for show-level browsing.

One important limitation concerns deletion. Greening reports that Prime sometimes returns success for old entries but leaves them in the history. His proposed safeguard is to reread history back to the oldest deleted day and restore any surviving entries to the local view. In that design, a successful response is not treated as sufficient evidence that the remote history changed.

Netflix requires a different integration

For Netflix, Greening describes reading viewing activity through the pathEvaluator endpoint, which lists individual episodes or movies. He says page zero always contains 20 rows and the response includes vhSize, the total row count.

Deletion uses the HideTitleViewing GraphQL mutation as a persisted query, submitted by query ID rather than query text. The hideAllEpisodes option can remove a whole series, according to the post. Greening says the query ID must be extracted from the page’s scripts because it changes with Netflix builds. That is a concrete maintenance dependency, not just an initial setup step.

Why share a brief instead of code?

Greening argues that maintaining an open-source project would be harder than building this particular app. He expects Prime’s undocumented API to break and does not want the resulting support issues and feature requests.

The distinction matters for anyone following his example: the post supplies a starting specification, not a maintained integration or a guarantee that its endpoint details will remain valid. It also mentions BBC iPlayer’s continue-watching page, but does not provide an equivalent implementation recipe. The described behavior and API details are Greening’s account; they have not been independently tested for this draft.