Skip to content
Draft
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 53 additions & 0 deletions .github/release-notes-instructions.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
# Release-notes editorial instructions (customer-facing)

These instructions are read by the automated polish step in `changelog-sync.yml` and by any human
editing `src/pages/docs/release-notes.mdx`. The audience is **FutureAGI customers** — developers and
AI teams using the platform. Assume no knowledge of our internal codebase, tickets, or services.

## Your task
The mechanical importer (`scripts/changelog-from-release.mjs`) has just inserted a new
`## <version> (<date>)` section at the very top of `src/pages/docs/release-notes.mdx`, directly below
the `release-notes:insert-below` marker. Its bullets are raw conventional-commit subjects. **Rewrite
only that newest section** into customer-facing prose that matches the existing entries below it.

## Match the existing house style exactly
Look at the sections already on the page and mirror them:
- Bullets are `- **Benefit-led Title in Title Case:** one or two plain sentences.`
- Lead with what the user can now do or what now works — never the code mechanism.
- Phrase fixes as observed behavior, e.g. *"In some cases, X failed… This has been resolved."*
- Keep the existing MDX wrapper markup and the `Features` / `Bugs/Improvements` / `Breaking Changes`
subsection headings (only include a subsection that has content).

## Cluster (the main job)
- Merge every bullet about the same feature or scope (the conventional-commit scope in parentheses)
into **one** bullet led by the combined user benefit. Collapse exact duplicates (the same subject
may appear twice from two commits — that is one bullet, not two).
- A dozen raw commits should become a handful of benefit statements.

## Drop or hide internal-only changes
- Drop changes a customer cannot observe: tests, CI, refactors, chores, dependency bumps, and
internal-only performance/index work. If a change has a user-visible effect, keep it and describe
the effect, not the mechanism.
- Never expose internal ticket IDs, internal service or module names, database/index/SQL details, or
source file paths.

## Accuracy (critical — these are published to customers)
- Describe only what the source bullets state. **Never invent** API names, flags, endpoints, model
IDs, or version numbers not present in the source. When unsure, omit rather than guess.
- Do not assert capabilities, limits, or guarantees the source does not state.

## Breaking changes
- Always in their own **Breaking Changes** subsection, each with a one-line migration note.

## Leave these aids in the file for the human reviewer
- If unsure how to phrase an item or which cluster it belongs to, keep the original bullet and append
an inline `{/* REVIEW: ... */}` comment explaining the doubt.
- At the end of the section, add one `{/* SKIPPED: ... */}` comment listing what you dropped and why,
so the reviewer can veto a drop.

## Hard constraints
- Edit **only** the newest `## <version>` section. Do not touch the marker, any older section, or any
file other than `src/pages/docs/release-notes.mdx`.
- Preserve valid MDX: escape stray `<`, `{`, `}` in prose. Keep any commit/PR links from the source
on the corresponding polished bullet where natural (the importer may strip them — do not fabricate).
- Leave your edits in the working tree; do not commit, push, or open a PR.
Loading