preview

Preview Pending Changeset Output

Render a preview of what the CHANGELOG entries would look like if all pending changesets were released now.

Step 1: Load Format Rules

Load the format skill via the Skill tool to get the authoritative list of the 13 section categories and their priority order. Use the priority table from that skill for all ordering decisions in the steps below.

Step 2: Find Changeset Files

List all .md files in the .changeset/ directory:

• Exclude any file named README.md
• If no files are found, report "No pending changesets" and stop

Step 3: Read and Parse Each File

For each discovered file, read its full contents and extract:

1. Frontmatter — parse all package-to-bump-type mappings from the YAML block between the --- delimiters (e.g., "@savvy-web/foo": minor)
2. Sections — parse the body content into a map of ## Heading → content blocks (paragraphs, lists, sub-headings, code fences, etc.)

If a file cannot be read or its frontmatter cannot be parsed, skip it and note it in the output as (unreadable).

Step 4: Group by Package

Build a per-package index. For each package that appears in any changeset frontmatter:

• Collect all bump types listed for that package across all changesets
• Collect all section content contributed by each changeset that includes that package

Step 5: Generate Per-Package Preview

For each package, compute its preview entry:

5a: Determine Bump Type

Select the highest bump level listed for that package across all changesets, using the precedence: major > minor > patch.

5b: Merge Sections

Gather all content contributed under each section heading across all changesets for this package. For each of the 13 known section categories that has any content:

• Collect all list items contributed under that heading from every changeset
• Collect all paragraphs, sub-headings, and code fences contributed under that heading from every changeset
• Deduplicate identical list items (exact string match, trimmed)
• Preserve non-list content (paragraphs, sub-headings, code blocks) in the order they appear across changesets, without deduplication

5c: Order Sections

Sort the merged sections by their priority number from the format category table (Breaking Changes = 1 first, Other = 13 last). Omit any category that has no content.

Step 6: Render the Preview

Output the preview in markdown, one block per package. Sort packages alphabetically by package name.

## @savvy-web/package-name (minor)

### Features

- item from changeset 1
- item from changeset 2

### Bug Fixes

- item from changeset 3

Format rules for the rendered output:

• Package header uses ## with the package name and bump type in parentheses
• Section headings use ###
• List items are rendered as - item
• Non-list content (paragraphs, sub-headings, code blocks) is reproduced as-is beneath the section heading, before any list items from the same section
• If a package has no parseable body content at all, note (no content) under its header

Step 7: Add Preview Notice

After all package blocks, append this note:

Preview only. The actual CHANGELOG is generated by the @savvy-web/changesets formatter, which performs additional processing: contributor footnotes, issue and PR link refs, dependency table rendering, and remark post-transforms. Content and ordering may differ slightly from the final output.

Output Guidance

• Do not validate changeset format or report rule violations — use /changesets:check for that
• Do not modify any files — this is a read-only preview
• If only one changeset exists for a package, the preview is effectively that changeset's content rendered in section order
• If a section heading in a changeset does not match any of the 13 known categories, include it at the end under its original heading text, as if it had priority 13