5 changes since v2.1.312, landing in 21 of 22 prompts.
Every prompt v2.1.314 sent, compared word for word with the capture before it. Each card is one edit, with every prompt it reached; the table after them shows where each one landed.
What changed 5
1
System prompt · under “Context management”
New rule bars recreating specific existing games, allows original ones
A token budget line under Context management is replaced with a game-building rule. Claude is now told it may build original games, games in a genre, and games from the user's own designs and assets, but not a recreation of a specific existing game, even if renamed or labeled original. The change landed in 18 of 22 captures.
+36−5 words
Copy 1 of 3 · 16 prompts, 4 model strings · under “Context management”
+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.
Copy 2 of 3 · 1 prompt, 4 model strings · under “Context management”
+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.
Copy 3 of 3 · 1 prompt, 4 model strings · under “Context management”
+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.
+EndConversation (deferred tool): use only for sustained user abuse directed at the assistant, or when the user explicitly asks to see it demonstrated. Load the full guidance via ToolSearch("select:EndConversation") before using it.
2
System prompt · under “Context management”
Remaining-token budget line dropped from system prompt
The line stating 15,000,000 tokens left is removed from the system prompt, so Claude no longer sees a total token budget there. This landed in only 2 of 22 captures, so it likely applies to a specific setup.
EndConversation deferred tool added, limited to sustained abuse or demos
A new line under Context management introduces EndConversation as a deferred tool. Claude may use it only for sustained user abuse aimed at the assistant or when the user explicitly asks to see it demonstrated, and it must first load the full guidance with ToolSearch("select:EndConversation"). The line appeared in only 1 of 23 captures, so it may be specific to one setup.
Behind a feature flag
+34−0 words
+EndConversation (deferred tool): use only for sustained user abuse directed at the assistant, or when the user explicitly asks to see it demonstrated. Load the full guidance via ToolSearch("select:EndConversation") before using it.
4
Appended context (skills, environment) · at “<total_tokens>15000000 tokens left</total_tokens>”
Total token budget notice removed from appended context
The appended context no longer includes the line stating 15,000,000 tokens left, so Claude is not told a remaining token budget there. This appeared in only 2 of 22 captures, suggesting it is specific to certain configurations.
+0−5 words
Copy 1 of 2 · 13 prompts, 6 model strings · at “<total_tokens>15000000 tokens left</total_tokens>”
Appended context (skills, environment) · at “EndConversation”
EndConversation entry added to appended context in one capture
A single line reading "EndConversation" now appears in the appended skills and environment context. It landed in only 1 of 13 captures, so it is likely tied to one setup, and the diff shows no description of what it does.
Behind a feature flag
+1−0 words
+EndConversation
Where each change landed
Every prompt the release sent, one row each. A number is a change from the list above; click it to jump there. Rows with more than one model string count them, because each model gets its own copy of the prompt.
