8 changes since v2.1.322, landing in 20 of 20 prompts.
Every prompt v2.1.323 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 8
1
Workflow tool description · line 4
Ultracode confirmation now called a message from the system
The two Workflow trigger conditions for ultracode now say a "message from the system" confirms the keyword or session mode, where they used to name a system-reminder. The conditions themselves are unchanged.
+8−4 words
~- The user included the keyword "ultracode" in their prompt (you'll see a system-remindermessage from the system confirming it).
~- Ultracode is on for the session (a system-remindermessage from the system confirms it) — see **Ultracode** in the workflow authoring reference.
2
ToolSearch tool description · at the opening
Deferred tool names described as arriving in a system message
ToolSearch now says deferred tools appear by name "in a message from the system" rather than in <system-reminder> messages. How deferred tools are fetched and then called is unchanged.
+5−3 words
~Deferred tools appear by name in <system-reminder> messagesa message from the system. Until fetched, only the name is known — there is no parameter schema, so the tool cannot be invoked. This tool takes a query, matches it against the deferred tool list, and returns the matched tools' complete JSONSchema definitions inside a <functions> block. Once a tool's schema appears in that result, it is callable exactly like any tool defined at the top of the prompt.
3
Skill tool description · at the opening
Skill listing now described as coming from the system
The Skill tool now says available skills appear "in a listing from the system" instead of a system-reminder listing. The guidance on when and how to invoke skills is the same.
+4−3 words
~A skill is a packaged set of instructions the user or project has set up for a particular kind of task (deploy steps, a review checklist, a repo-specific workflow). Available skills appear in a system-reminder listinglisting from the system with one-line descriptions. When the task at hand is one a listed skill covers, call this tool first — the skill's instructions load into the turn for you to follow in place of your default approach; some skills instead run in a subagent and return the finished result. A skill that runs in the background returns only the agent's name — its result arrives later as a task notification, so don't wait on it or invoke it again in the meantime. Users may also ask for one by name (`/<name>`, or "slash command"); that's a request to invoke it.
4
Agent tool description · at the opening
Agent types listed in messages from the system, not system-reminders
The Agent tool now says available agent types are listed in "messages from the system" rather than in <system-reminder> messages. This is a wording change with no change in behaviour.
+4−3 words
~Available agent types are listed in <system-reminder> messagesmessages from the system in the conversation.
5
Agent tool description · under “Writing the prompt”
Ship-audit example prompt says feature flag instead of GrowthBook gate
In the example subagent prompt for auditing a branch before shipping, "the GrowthBook gate" is replaced with the generic "the feature flag". The example otherwise reads the same. It appeared in only 6 of 17 captures.
+2−2 words
~ prompt: "Audit what's left before this branch can ship. Check: uncommitted changes, commits ahead of main, whether tests exist, whether the GrowthBook gatefeature flag is wired up, whether CI-relevant files changed. Report a punch list — done vs. missing. Under 200 words."
6
Agent tool description · under “Writing the prompt”
Audit example reply says feature flag instead of GrowthBook gate
The example assistant reply summarizing the audit now lists a "feature flag" blocker instead of a "GrowthBook gate" blocker. This matches the edit to the example prompt, and the meaning is the same. It appeared in only 6 of 17 captures.
+2−2 words
~assistant: Audit's back. Three blockers: no tests for the new prompt path, GrowthBook gatefeature flag wired but not in build_flags.yaml, and one uncommitted file.
7
Bash tool description · under “Committing changes with git”
Git commit rule adds .claude/settings.local.json as a likely secret
The Bash git guidance now names .claude/settings.local.json alongside .env and credentials.json as files Claude should not commit. Claude should warn the user if they ask to commit it. This appeared in 6 of 17 captures.
+4−0 words
~ - Do not commit files that likely contain secrets (.env, credentials.json, .claude/settings.local.json, etc). Warn the user if they specifically request to commit those files
8
Appended context (skills, environment) · entries “init” and “security-review”
init and security-review skills get one-line descriptions
In the appended skills listing, init and security-review now carry descriptions. init creates a CLAUDE.md with codebase documentation, and security-review reviews the pending changes on the current branch. Before, only their names were listed. This appeared in 6 of 17 captures.
Only for default
+24−3 words
−- init
−- security-review
+- init: Initialize a new CLAUDE.md file with codebase documentation
+- security-review: Complete a security review of the pending changes on the current branch
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.
