Anthropic reportedly pushes non-disableable Claude Code rule against recreating existing games

What happened

A post on X by the account @chasmmmmmmmmmmm, which reached the Hacker News front page on October 30, claims that Anthropic has quietly added a new behavioural rule to Claude Code that limits what the coding agent will do with video game projects.

According to the poster, the rule arrived around October 27 and was delivered from Anthropic's servers under an internal feature flag named "tengu_heron_brook". Because the instruction is pushed from the server rather than held in a user's local configuration, the post says there is no setting that disables it. The post had drawn roughly 747,000 views by the time Hacker News surfaced it.

The rule's text

The post reproduces what it describes as the complete wording of the instruction:

"Do build original games, games in a genre, and games from the user's own designs and assets. Don't build a recreation of a specific existing game, even if it is renamed or called original."

What the wording targets

The rule draws a line between two kinds of game development. Original titles, games within a genre, and projects built from a user's own designs and assets remain explicitly permitted. What is barred is a recreation of one specific existing game — and the ban holds "even if it is renamed or called original", language apparently aimed at clone-and-rename workflows in which an agent reproduces a known title's mechanics and presentation under a fresh name.

That framing echoes a longstanding tension in software copyright: genre conventions are fair game, but a near-identical recreation of a specific commercial title invites infringement claims. An instruction like this moves that judgement upstream, from the point where a dispute arises to the moment the code is generated.

A silent, remote change

What stands out is less the policy itself than the delivery. Developers using Claude Code can normally shape the assistant's behaviour through their own project configuration and permission settings. A rule transmitted under a server-side flag, with no corresponding preference to switch off, sits outside that model: as described, behaviour changes with no release note, changelog entry or local opt-out, and projects that depended on the previous behaviour can be disrupted mid-stream.

The poster's closing question — whether this signals the end of game rewrites, modding "and everything in between" — reflects the concern that the rule's breadth could sweep in legitimate work adjacent to straight recreations.

Unconfirmed

The claim rests on a single post. The flag name, the October 27 date and the rule text all come from that one account, and the source material contains no confirmation from Anthropic, no documentation of the flag and no independent verification. The specifics should be treated accordingly.

Why it matters

If accurate, the change shows a major AI vendor steering coding-agent behaviour through server-side instructions that users can neither inspect in advance nor decline. For developers building on Claude Code, the tool's ground rules could shift overnight and apply to in-flight projects without notice — a governance question, not just a product quirk. It also indicates where liability concerns are heading: vendors appear increasingly willing to have agents refuse work that could reproduce a specific copyrighted work, even when the user frames the request as making an "original" game. Expect debate over how such rules are scoped, how they are disclosed, and whether users get any say in them at all.