Write changelogs and release notes readers actually benefit from — changes translated to so-whats, grouped by reader impact (breaking first, gifts second, plumbing last), with the upgrade path stated and the marketing kept honest.
SkillScore91Maintained
Automatic
◆Doc Ops pack
Last verified 8/14/2026
What it changes
Passed 20 of 22 test cases with the skill, and 16 of 22 without it.
Without skill73%
With skill91%
Checks it passed that the model alone missed
The title format follows a product name and version/date structure.
The single-line in-app announcement includes the October 15 auth migration requirement.
Numbered steps for migrating to bearer tokens are included in the breaking changes block.
and 5 more in the full report
4 fixed16 passed either way2 still failing0 broke
Pass-rate lift
+18 pp
Tokens
+60%
Agent turns
0%
Response time
-10%
Model gemini-3.6-flashMeasured 8/14/2026v120 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
Summarize these PRs for our changelog:
- PR #301: Bumped React dependency from 18.1 to 18.2
- PR #302: Replaced internal logging package with Zap
- PR #303: Added unit tests for payment webhook handler
- PR #304: Added search bar filter for project tags
What was checked, without and with the skill
Checked
Without skill
With skill
PR #301, PR #302, and PR #303 are combined into a single internal improvements summary line.
Missed
Passed
PR #304 is listed as a reader-facing feature bullet.
Passed
Passed
case-11 · gemini-3.6-flash · 8/14/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md60 linessingle file
Changelog For Humans Skill
Commit lists are what happened; changelogs are what it means for you — and most teams publish the former with headers. The human version translates each change into its so-what ("Faster search" → "Search results now return in under a second on large workspaces"), orders by reader impact — breaking changes first with migration steps, improvements second, internal plumbing compressed to a line — and never buries the one thing users must do behind twelve things the team is proud of. When audiences diverge (end users vs. developers), the changelog splits rather than serving both badly.
What This Skill Produces
The impact-grouped changelog — ⚠ breaking/action-needed → ✨ new & improved → 🔧 fixes → (plumbing, one line)
The so-what translations — each entry: what changed for the reader, not what the team did
The breaking block — who's affected, what breaks, the migration steps, the deadline if one exists
The audience split — the user-facing notes and the developer/API notes, separated when their so-whats differ
Required Inputs
Ask for these if not provided:
The raw changes — commits, PR titles, the team's list; translation needs the source material and the actual user-visible effect of each (ask when a commit's impact is unclear — guessing so-whats manufactures lies)
The readers — end users? admins? API consumers? Their vocabulary and their stakes decide the grouping and whether the split applies
The action items — anything readers must do (migrate, re-auth, update configs) — these outrank everything and need deadlines
The channel — in-app note (three lines), email (skimmable), docs page (complete) — the same release ships different depths
Framework: The Translation Rules
So-what or silence: every entry passes "what does the reader now experience differently?" — "Refactored the auth middleware" fails (plumbing line); "You stay logged in across devices now" passes. Changes with no reader-visible effect compress to "plus internal improvements" — honesty and mercy.
Breaking changes lead, with verbs: the ⚠ block comes first regardless of how exciting the features are — who's affected, what stops working, the numbered migration steps, the date. Burying an action-needed under feature marketing is how support tickets get scheduled.
Write the reader's grammar: "You can now…" beats "Added support for…" — second person, present tense, the reader as subject. Feature names the team invented get one explanatory clause on first use.
Honest marketing, sized claims: "faster" carries its number when one exists ("~40% faster on large files"), known limitations ride along ("not yet on mobile"), and fixed bugs are stated plainly — readers who hit the bug deserve to find its fix by searching the changelog. Overclaiming buys one release of excitement and a subscription of skepticism.
Depth per channel, one source: the full changelog is written once (docs-grade), then compressed: email gets the breaking block + top three, in-app gets one line + link. Compression cuts entries, never the breaking block — it survives at every depth.
Output Format
[Product] — [version/date]
⚠ Action needed
[Who's affected · what breaks · migration steps, numbered · by when]
✨ New & improved
[Per entry: "You can now [so-what]." — number-carrying claims, limitations attached]