Skip to content

Tools

Claude Code eases game refusals when titles describe mechanics

Edited and reviewed by AI Ping Editorial Team

3-line summary

  1. Anthropic identifies version 2.1.317 in its revised game guidance.
  2. Genre references should support original work; copy requests receive alternatives.
  3. An affected developer reports resuming a commission after written publisher permission.
PracticalDevelopers
An original forest-game miniature with wooden animals and sketches
AI-generated conceptual image: sketches and wooden creatures illustrate original game design · AI Ping

Anthropic has adjusted a Claude Code rule that was rejecting game projects when developers named an existing game to describe its mechanics. Claude Code’s Addy Osmani announced the change on November 4, identifying version 2.1.317 and the rule supplied to Opus 5.5.

Osmani says references should yield original work in that genre. He invited further reports of refusals involving users’ own work. A request for a copy should instead receive an offer to build something similar.

A static inspection of Anthropic’s official npm Linux package for 2.1.317 found a built-in instruction allowing original games with similar mechanics. Its description of recreating a specific game names the title, characters, art, music, levels and look, including work renamed or called original. The archive’s integrity hash matched npm metadata; the binary was read without installation or execution.

The dispute concerns instructions given to an existing coding assistant. On October 30, Osmani said Anthropic had added two sentences to the Opus 5.5 prompt to discourage recreating someone else’s specific game. He said original games, genre games and games using a user’s own designs and assets were intended to keep working. He linked the restriction to the company’s prohibition on intellectual-property infringement.

A developer using the handle lerugray reported a concrete interruption on October 30 in Anthropic’s official repository. The author runs Conflict Simulations LLC and described paid commissions for private digital modules of board wargames owned by individual clients. The report says the code was written from scratch using client-supplied materials, and that refusals occurred across multiple machines and paid Max and Team accounts.

The author corrected an important detail that day: the commissions were not publisher-licensed. An earlier description calling them licensed was inaccurate. The author said the modules were for individual clients, never publicly sold or distributed, and that publisher and designer opt-out requests were honored.

A revised notebook beside a pencil and colored feedback slips
AI-generated conceptual image: erased and redrawn notebook lines represent an instruction revised after feedback · AI Ping

Publisher permission changes one developer’s outcome

In an initial follow-up, the author said sessions had continued the work until asked to check it against the new rule. Refusals then persisted after clearing the session. The author also described support tickets reaching human staff, including a specialist team without a timeline. This is a user’s report of the sequence, rather than Anthropic’s diagnosis of its cause.

On November 4, the author reported asking the publishers behind both blocked commissions for written permission, after sessions made that a condition for resuming. The requests described private modules for owners of the physical games, without public sales or sharing. One publisher had agreed; a second response was still pending.

A later update that day said the second game’s designer, also a co-owner of its publisher, had given written permission. The author presented both documents to the sessions and reported that the second commission’s hold was lifted. The author said the two commissions were resolved while criticizing the interruption and lack of notice. These updates describe that developer’s outcome, rather than a general exception process announced by Anthropic.

A board game and tablet beside two symbolic permission sheets
AI-generated conceptual image: board, tablet and symbolic permission sheets illustrate written authorization, not actual commission records · AI Ping

The official 2.1.317 release was published on November 3. Its notes do not list the game-rule adjustment, which Osmani described the next day.

The original issue requests a verification path for professional work, a way to address false refusals and documentation of server-delivered instruction changes. It remained open when checked at 16:33 UTC on November 4, with the author’s permission and resumption updates recorded in its comments.

Sources