Installable skillApache-2.0

Changelog Writer

Create user-focused, SEO-optimized changelog entries for software releases. Use when writing release notes, version updates, product changelogs, or "what's new" documentation for developer tools.

BY David-Li0406UPDATED Oct 27, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: changelog-writer
description: Create user-focused, SEO-optimized changelog entries for software releases. Use when writing release notes, version updates, product changelogs, or "what's new" documentation for developer tools.
allowed-tools: Read, Grep, Glob, Write, Edit
---

# Changelog Writer

Create clear, accurate changelog entries that help developers understand what's new in Lightfast releases.

## Critical: Fact-Check First

Before writing anything, verify against `docs/architecture/implementation-status/README.md`:

1. **Check implementation status** to verify:
   - What's actually completed vs planned
   - Current limitations and known gaps
   - Technical accuracy of claims

2. **Never oversell:**
   - Use specific names: "GitHub File Sync (File Contents)" not "GitHub Integration"
   - Disclose limitations: "Currently supports X; Y coming in vZ"
   - Be honest about conditionals: "when 3+ customers request"

3. **Verify every claim:**
   - If you cite a number, confirm it's in implementation docs
   - If you mention a feature, confirm it exists in production
   - When uncertain, ask for clarification

## Writing Guidelines

1. **Concise & scannable**: 1-3 sentences per feature (Cursor-style brevity)
2. **Lead with benefit**: Start with what users can do, then how
3. **Be transparent**: Mention beta status, rollout timelines, limitations
4. **User-focused but technical**: Balance benefits with specifics developers need
5. **Active voice**: "You can now..." not "Users are able to..."
6. **No emoji**: Professional tone
7. **Specific examples**: Include config snippets, API calls
8. **SEO-conscious**: Use target keywords naturally
9. **AEO-conscious**: Write `tldr` for AI citation engines, `excerpt` for listings
10. **FAQ quality**: Questions must match real search queries, answers must be complete

## Workflow

1. **Gather input**: PR numbers, URLs, or manual change list
2. **Read implementation status** for fact-checking
3. **Draft following** [templates](resources/templates.md)
4. **Cross-check claims** against implementation reality
5. **Add SEO elements** per [seo-requirements](resources/seo-requirements.md)
6. **Review with** [checklist](resources/checklist.md)

## Quick Reference

### Do
- "GitHub File Sync (File Contents)" with limitations disclosed
- "When 3+ customers request: Linear integration"
- Include code examples for every major feature
- Link to 3-5 related docs

### Don't
- "GitHub Integration" (vague - what does it cover?)
- "Coming soon: Linear, Notion, Slack!" (when at 0%)
- Long paragraphs (keep to 1-3 sentences per feature)
- Claims without verification

## Output

Save drafts to: `thoughts/changelog/{title-slug}-{YYYYMMDD-HHMMSS}.md`

### Required Frontmatter Fields

Every draft MUST include:
- `title`, `slug`, `publishedAt` (core)
- `excerpt`, `tldr` (AEO)
- `seo.metaDescription`, `seo.focusKeyword` (SEO)
- `_internal.status`, `_internal.source_prs` (traceability)

### Slug Format

Always use: `0-<version>-lightfast-<feature-slug>`

Examples:
- `0-1-lightfast-github-file-sync-semantic-search`
- `0-2-lightfast-pr-metadata-linear-integration`

Recommended:
- `seo.secondaryKeyword`, `seo.faq[]` (enhanced SEO)

The frontmatter structure maps directly to `ChangelogEntryInput` type. Use `/publish_changelog` to publish drafts to BaseHub.

See `resources/templates.md` for complete frontmatter template.

## Resources

- [Document Templates](resources/templates.md)
- [SEO Requirements](resources/seo-requirements.md)
- [Examples](resources/examples.md)
- [Pre-Publish Checklist](resources/checklist.md)
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.