[eugenia-scandit]

Two small follow-ups to #430, split out so that PR could keep the approval it had already earned. Both were found by a review round that ran after #430 was approved; neither is urgent, and neither changes behaviour today.

Base note: this targets fix/framework-frontmatter-ssot so the diff shows only these two files. GitHub retargets it to main automatically once #430 merges.

1. shQuiet() is dead, and a comment still points at it

#430's last commit replaced frontmatterOnly's only shQuiet call with execFileSync("git", [...]) — deliberately, to stop interpolating a git-derived path into a shell string — but left the function defined and left the call-site comment reading "see the note on shQuiet".

The cost is not the dead function, it is the instruction attached to it: the next reader is told stderr suppression comes from shQuiet, goes looking, finds nothing calling it, and if they "restore" the call they reintroduce exactly the class #430 removed.

Deleted. The comment now says what actually suppresses the noise — stdio: ["ignore", "pipe", "ignore"] — and why there is noise to suppress: a file the PR adds has no base blob, so git prints fatal: path ... exists on disk, but not in ... once per new page. That failure is expected and handled (the catch treats the file as body-changed, which is right for a new file), but forty such lines ahead of the gate's own output reads as a crashed gate rather than as forty new pages.

2. display is a map key in four places and nothing pinned its uniqueness

slug uniqueness was asserted. display was not, despite being keyed off in:

• FRAMEWORK_URL_PATH and FRAMEWORK_SLUG (SkillsCallout)
• FRAMEWORK_DISPLAY_TO_SLUG (utils/frameworks.ts)
• the features.json / products.json / skills.json cross-check in verify-frameworks.cjs

The field's own doc comment already says "Used as a map key by SkillsCallout, so it is load-bearing."

A duplicate is easy to introduce — a copy-pasted entry whose display was not updated. Object.fromEntries silently keeps only the last one, so SkillsCallout would build the other framework's /sdks/<routeSegment>/agent-skills link and report the wrong framework to analytics. yarn verify:frameworks stays green throughout, because it only ever asks whether a display is in the registry, never whether it appears once.

Verification

• yarn test:frameworks → 17 passed (16 + the new row)
• yarn test:docs-gate → 37 passed
• yarn verify:frameworks → OK
• npx tsc --noEmit → clean
• Mutation-tested: changing linux's display to "Web" fails the new assertion. It bites.

🤖 Generated with Claude Code

[moritzhartmeier]

Heads-up: I force-pushed this branch to clear the conflicts. Previous head was 57ee573a4 if you need to recover anything.

Cause: #430 was squash-merged, so main got the whole thing as a single commit (804180a1e) while this branch still carried the 30 original commits. Retargeted from fix/framework-frontmatter-ssot to main, it was replaying work already in main — hence 136 changed files and the conflicts. Nothing was actually wrong with your change.

What I did: cherry-picked your 57ee573a4 onto current main and dropped the inherited history. It applied cleanly, no conflict resolution, and your authorship is intact:

git checkout -b <branch> origin/main
git cherry-pick 57ee573a4

This PR is now 1 commit, 2 files, +29 -14 — just the change you meant it to be, and MERGEABLE.

Verified on the rebased tree before pushing: no shQuiet references remain anywhere under scripts/, test:frameworks 17 passed, test:docs-gate 37 passed. I also checked the new uniqueness test earns its place rather than trusting the comment — injecting a duplicate display (Cordova -> iOS) fails it loudly, and FRAMEWORK_DISPLAY_TO_SLUG really is an Object.fromEntries keyed on display, so the silent last-wins overwrite you describe is real.

If you had local work on top of the old head, git fetch && git reset --hard origin/<branch> and re-apply.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-22 10:49 UTC