What's New in Claude Code

Provide a curated summary of new Claude Code features. Supports two modes:

• Automatic mode (default): Show changes since the last time the user checked, tracked via a cache file.
• Explicit version mode: The user specifies a starting version (e.g., "what's changed since 1.0.30?"). Use the version they provide as the starting point.

Determine the version range, fetch the changelog, identify notable features, then spawn parallel agents — one per feature — for research and article writing.

Workflow

Execute these steps in order:

Step 1: Determine Versions

First, check if the user specified a starting version in their request (e.g., "since version 1.0.30", "since 1.0.30", "changes after 1.0.30"). If so, use that as the starting version and skip reading the cache file.

Otherwise, run these two Bash commands in parallel:

1. claude --version — to get the currently installed Claude Code version
2. cat ~/.claude/whats-new-in-claude-last-version 2>/dev/null — to get the last-checked version

Always run claude --version to get the current version.

Parse the version numbers from the output.

Determining the starting version (in priority order):

1. User specified a version: Use exactly what they said.
2. Cache file exists and has content: Use its contents as the starting version.
3. No cache file / first run: The starting version will be determined from the changelog in Step 3 (last 10 versions).

Step 2: Check If Already Current

If the starting version equals the current version, inform the user there are no new changes since that version. Do not update the cache file. Stop here.

Skip this check if the user explicitly provided a version — they may want to re-read a range they've already seen.

Step 3: Fetch and Filter the Changelog

Fetch the Claude Code changelog using WebFetch:

URL: https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md
Prompt: Extract all version headings and their content between version [current] and version [last-checked]. If this is a first run (no last-checked version), extract the last 10 version sections. Return the raw changelog entries for these versions.

Use the raw GitHub URL for clean markdown without HTML wrapper noise.

If the last-checked version does not appear in the changelog, fall back to the last 10 versions.

Step 4: Identify Notable Features

Review the extracted changelog entries and build a list of notable features. Each feature should be a distinct capability or improvement worth writing about.

Include: New features, significant improvements, new commands, new integrations, breaking changes, deprecations.

Exclude: Bug fixes, patch-level fixes, minor tweaks, dependency updates, internal refactors.

For each feature, note:

• A short title
• The version it appeared in
• The relevant changelog text for that feature

Aim to identify all features worth covering individually. Do not group unrelated features together — each gets its own agent.

Step 5: Spawn Parallel Research Agents

This is the critical parallelization step. Use the Task tool to spawn one whats-new-in-claude:changelog-researcher agent per feature, all in a single message so they run in parallel. Use model: haiku for each agent to keep things fast and cheap.

For each agent, provide a prompt like:

Research this Claude Code feature and write a detailed article about it.

Feature: [title]
Version: [version]
Changelog entry: [the raw changelog text for this feature]

Search the web for additional context on anthropic.com and code.claude.com, then write a 2-3 paragraph article.

Launch ALL agents in a single message — do not wait for one to finish before starting the next.

Step 6: Collect and Present Results

As agents return, collect their articles. Once all agents have completed:

1. If every agent returned NO_NEW_FEATURES, tell the user: "No notable new features since version [starting-version] — just bug fixes and stability improvements."
2. Otherwise, present the collected articles to the user as a single cohesive document:
   • Add an intro line like "Here's what's new in Claude Code since version [starting-version]:"
   • Order articles from most recent version to oldest
   • Add --- separators between articles
   • If there are many features, add grouping headers (e.g., "Developer Experience", "Performance", "New Integrations") but keep every article intact — do not summarize or compress them

Step 7: Update the Cache

After presenting the summary, use Bash to write the current version to the cache file:

echo "[current-version]" > ~/.claude/whats-new-in-claude-last-version

Edge Cases

• Changelog fetch fails: Inform the user that the changelog could not be retrieved. Suggest checking their network connection. Do not update the cache file.
• Version not found in changelog: Fall back to showing the last 10 versions.
• Very large gap (>20 versions): Identify the most significant features (cap at ~15 agents to avoid excessive parallelism). Mention to the user that many versions were covered and some minor items were omitted.