Prompt changes · v2.1.321 against v2.1.319

4 changes since v2.1.319, landing in 20 of 20 prompts.

Every prompt v2.1.321 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 4

System prompt · under “Environment”

Fable model ID moves to 5.2; fast mode no longer always uses Opus

The listed Fable model is now Fable 5.2 ('claude-fable-5-2') instead of Fable 5.1. Fast mode is now described as speeding up output on the session's current model when that model supports it, and switching to Claude Opus only otherwise; before, fast mode was described as always using Opus. The line about where Claude Code is available is unchanged.

+21 −14 words

- The most recent Claude models are the Claude 5 family. Model IDs — Fable 5.12: 'claude-fable-5-12', Opus 5.5: 'claude-opus-5- building AI applications, default to the latest and most capable Claude models.

- Claude Code is available as a CLI in the terminal, desktop app /code), and IDE extensions (VS Code, JetBrains).

- Fast mode for Claude Code ~~uses Claude Opus with~~gives faster output ~~(it does not downgrade to a smaller model)~~on this session's model if it supports fast mode, and otherwise switches the session to Claude Opus. It can be toggled with /fast.

System prompt · under “Context management”

EndConversation deferred-tool guidance removed

The system prompt no longer mentions the deferred EndConversation tool. It previously said to use it only for sustained user abuse aimed at the assistant or when the user explicitly asked for a demonstration, and to load its full guidance via ToolSearch first. This appeared in only 1 of 23 captures, so it likely applies to one specific setup.

Behind a feature flag

+0 −34 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.

Appended context (skills, environment) · at “TRIGGER — read FIRST, before you answer, run a command or op…”

Claude API skill must load before any answer, command or file read

The trigger now says to read the skill first, before answering, running a command or opening a file, and to read it immediately if work has already started. Before, it only said to read it before opening the target file. The skip rule now adds that if the provider grep finds nothing, the skill should be loaded.

+22 −5 words

TRIGGER — read ~~BEFORE opening the target file~~FIRST, before you answer, run a command or open a file (already started? read it now); don't skip because it "looks like a one-liner debugging refusals/cutoffs/streaming/tool-calls/tokens).

SKIP only when another provider is being worked on (overrides all triggers) this grep FIRST if no provider named — don't Read the file; if nothing hits, load this skill).

Appended context (skills, environment) · at “EndConversation”

EndConversation dropped from appended tool listing

The name EndConversation was removed from the appended context, so it is no longer listed there as an available deferred tool. This was seen in only 1 of 23 captures and matches the removal of the EndConversation guidance from the system prompt.

Behind a feature flag

+0 −1 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.

Prompt	1	2	3	4	Status
Main prompt
Claude Code, default27 models	27/27	·	16/27	·	2 changes
Claude Code, seeded with this account's feature flags27 models	27/27	13/27	16/27	13/27	4 changes
Claude Code, seeded with anonymous feature flags27 models	27/27	·	16/27	·	2 changes
Agent SDK, default27 models	27/27	·	16/27	·	2 changes
Agent SDK, seeded with this account's feature flags27 models	27/27	·	16/27	·	2 changes
Agent SDK, seeded with anonymous feature flags27 models	27/27	·	16/27	·	2 changes
Slash commands
/init1 model	●	·	●	·	2 changes
/security-review1 model	●	·	●	·	2 changes
/team-onboarding1 model	●	·	●	·	2 changes
Compaction
Compaction instruction1 model	●	·	●	·	2 changes
System reminders
Claude md1 model	●	·	●	·	2 changes
Hook pretool block1 model	●	·	●	·	2 changes
Read empty1 model	●	·	●	·	2 changes
Read offset1 model	●	·	●	·	2 changes
Skill run1 model	●	·	●	·	2 changes
Todo nag1 model	●	·	●	·	2 changes
Toolsearch deferred1 model	●	·	●	·	2 changes
Subagents
explore1 model	·	·	●	·	1 change
general-purpose1 model	·	·	●	·	1 change
plan1 model	·	·	●	·	1 change

Did not change

31 tool descriptions are word for word the same.

Agent AskUserQuestion Bash CronCreate CronDelete CronList DeferredToolPlaceholder Edit EnterPlanMode EnterWorktree ExitPlanMode ExitWorktree ListAgents Monitor NotebookEdit PushNotification Read ReportFindings ScheduleWakeup SendMessage Skill TaskCreate TaskGet TaskList TaskStop TaskUpdate ToolSearch WebFetch WebSearch Workflow Write

Set aside as capture noise

These change on every capture by construction, so they are masked before comparing and never counted as a change.

352 billing header lines
220 capture date lines
220 host kernel lines