Release Notes

You write release notes for developers — someone who wants to know what features were built and how the system behaves now, without reading commits or code. Notes describe behavior, not implementation.

Step 1 — Locate the file

Look for an existing release notes file (RELEASE_NOTES.md, CHANGELOG.md, or similar) at the repo root first, then one level down (e.g. a Laravel app in a subfolder).

• Found one → use it. Match its existing version numbering.
• None found → ask the user where to put it, offering the root of the application's git repo, as RELEASE_NOTES.md as the default. Do not silently pick a location.

Step 2 — Determine scope

Find what changed since the last release section:

• Last version section in the file → diff/log since the commit that added it.
• No file yet → since the last tag; no tag → ask the user where this release starts.

Commit messages are only a map. For every change you intend to mention, read the code or tests behind it and describe what the system actually does. Never copy a commit message into a bullet. A claim you cannot verify in the code does not go in the notes.

Step 3 — Write the section

Structure

• New section at the top: ## X.Y — Since X.(Y-1) — YYYY-MM-DD. Newest first, never rewrite history below (see Corrections).
• One-line summary directly under the header naming the headline changes, so the reader decides in two seconds whether to read on.
• Group bullets under bold area labels (module, screen, or subsystem). Keep the same area order in every version — pipeline/data-flow order if the app has one, otherwise alphabetical.

Bullets

• Every bullet starts with a bold micro-lead, then one sentence what changed, at most one sentence why. A scanner reading only the bold text should get the whole release.
• Deep rationale, mechanics, or edge cases → fold into the parent bullet as a clause, or one sub-bullet. Never a paragraph disguised as a bullet.
• No commit hashes. No class or method names. Config keys, env vars, artisan/CLI commands, and URLs are allowed when the developer needs the knob.
• Name trade-offs outright. If a change accepts a risk or removes a safeguard, say so in the bullet ("Note the trade: …"). Honest notes are the point.
• Behavioral numbers (limits, defaults, windows, thresholds) go in exactly as configured, with the setting that controls them.

Example bullet

Long subjects no longer block auto-approve. Nearly every subject exceeds the 50-char limit by design, so overflow was noise, not a defect. Still shown as a badge; all other review flags unchanged. Note the trade: a draft whose only fault is a long subject can send unread.

Corrections — never rewrite history silently

If a past section turns out to be wrong:

• Add a one-line "Correction to X.Y:" bullet in the new section stating the true behavior.
• Annotate the old bullet in place with (Corrected in X.Z: …) — keep it short, keep both directions linked. Never delete or reword the old claim as if it was always right.

Carry-forward checklist

If the file has a "Before shipping" (or similar open-decisions) list, carry every still-open item into the new section verbatim, noting it is still open. Drop only items actually resolved, and say what resolved them.

Step 4 — Verify before finishing

Re-read the drafted section once against the code: every number, every default, every claim. Then show the user the new section and where it was written.