chore: enforce figma links in relevant prs

[ovitrif]

Closes #731

Description

This PR:

1. Adds a Design section to the PR template with Figma-link and explicit N/A guidance.
2. Aligns agent authoring and review rules for known and uncertain Figma frames.
3. Updates the release PR generator to include the required non-UI Design value.

Linked Issues/Tasks

• chore: enforce figma links in relevant prs #731

Design

N/A — no UI changes.

Preview

N/A — no user-visible changes.

QA Notes

Manual Tests

N/A

Automated Checks

• Local verification: policy-matrix inspection, release-body generation check, and git diff --check passed.

[jvsena42]

Both greptile findings and my parity note are closed — thanks.

The screens-map reference is gone repo-wide: git grep -E "screens-map|screens map|ltqvnKiejWj0JQiqtDf2JJ|figma\.com" at head returns zero hits, and the only Figma pointers are the three Bitkit - Handoff vNN mentions this PR adds. That was the right of the two options — the artifact didn't exist here. release.md:129-131 has ### Design, and the reviewer clauses now sit in pr.md:121-123 as well as AGENTS.md:347-349.

Minor, not worth changing on its own: the handoff page is named but no file key or URL travels with it, unlike the Android side which pins ltqvnKiejWj0JQiqtDf2JJ in docs/screens-map.md:3. Reaching the page still needs knowledge held outside the repo.

One thing Android fixed in its twin that this still carries — inline.

Cross-repo framing that's more useful than either note alone: Android's follow-up went the opposite direction. It closed the N/A loophole but narrowed its design case to mapped *Screen.kt only, dropping 44 sheets out of coverage. This repo's scope is the better one — it covers sheets, views, anything with a design — it just needs the exemption tightened. Between the two PRs, each has the half the other wants.

[jvsena42]

936b439e applies exactly the scoping I suggested, on both surfaces — AGENTS.md:348 and .agents/commands/pr.md:122 now read `N/A — no UI changes.` needs no review request; `N/A — no design available.` may receive the single advisory clarification. The self-declared branch is no longer blanket-exempt, which was the whole point. Thanks.

I also checked the two literal strings match byte-for-byte (em dash, trailing period) across every surface that emits or consumes them — .github/pull_request_template.md:15-16, AGENTS.md:343-344, .agents/commands/pr.md:117-118, and the release generator at .agents/commands/release.md:129-131. pr.md builds its body from the template and those two are the only PR-body generators in the tree, so no path can now emit a body without ### Design.

Nothing further from me.

[jvsena42]

Two LOWs inline. These are instruction documents — nobody executes them, but agents and humans follow them, so a wrong or unsatisfiable rule gets acted on repeatedly. That's the lens.

Verified sound:

• The missing docs/screens-map.md edit is correct, not a gap. The android twin (synonymdev/bitkit-android#1236) also edits its screens map, and I checked whether iOS should have an equivalent. It shouldn't: git grep -Ei "screens-map|screens map|ltqvnKiejWj0JQiqtDf2JJ|figma\.com" at head returns zero hits, there is no docs/ directory (only Docs/ with eight unrelated files), and no map-drift test exists. Nothing here was missed.
• Literal strings byte-identical across template :15/:16, AGENTS.md:343/:344/:348, pr.md:117/:118/:122, release.md:131. release.md's body order matches the template (Description → Design → Screenshot / Video), and :116 still reads the template first.
• No unbacked enforcement claim — AGENTS.md:349 and pr.md:123 say links never block approval or CI, and nothing in the repo reads the PR body.
• Rule reaches all agent surfaces — CLAUDE.md symlinks to AGENTS.md, .cursor/commands/pr.md symlinks to .agents/commands/pr.md.

Drift vs the android twin — mostly justified by the absence of a screens map, but three differences are not: android scopes the advisory request "per PR" where this doesn't scope it at all; android's pr.md:168 says links never block "approval, CI, PR creation, or review readiness" where every other surface on both repos says just "approval or CI"; and android deliberately drops "missing" from the uncertain-frame clause (its map makes a missing frame unreachable) where :345 here keeps "Missing or uncertain", which is right for this repo. Worth one pass to decide which wordings are meant to be identical.

[jvsena42]

Both fixes confirmed at e0f9bac2, and the drift closed. No findings.

Attribution first, because the diffstat looks alarming: git diff c88ad9d9 refs/remotes/pr/734 shows 23 files / +822 including PubkyService.swift and a pile of tests — but 19 of those are byte-identical to origin/master (git diff --quiet origin/master refs/remotes/pr/734 -- <path> exits 0 for every one), inherited from the 5ea3cf3d merge that brought in #697. git diff --stat origin/master refs/remotes/pr/734 returns exactly four files, all docs, +36/-0, matching gh pr diff --name-only. No source file is PR-authored and there is no evil merge — git diff origin/master…pr/734 touches zero non-doc files, so there was nothing for git show --cc to interrogate. A docs PR carrying no source changes is exactly right.

(a) resolved. pr.md:115 — "Include a ### Design section when the PR template contains a ### Design heading." Matches the file's own ### Preview gating at :188/:189, so the library-repo travelling case at :154 is handled. No behaviour change in this repo (the template does carry the heading) — this is purely portability.

(b) resolved, and the key is right. The file key landed in all three surfaces — pr.md:116, pull_request_template.md:14, AGENTS.md:342 — identical in each, as https://www.figma.com/design/ltqvnKiejWj0JQiqtDf2JJ/. Re-ran the grep at the new head: it now returns those three hits where it previously returned nothing, so the instruction is resolvable from inside the repo. I checked the key against android's docs/screens-map.md:3 — same file, and no typo drift between the three occurrences.

Drift, both items I flagged now fixed:

• Advisory-request scope: you adopted "per PR", matching android.
• Non-blocking wording: both repos now say "approval, CI, PR creation, or review readiness". git grep "approval or CI" returns zero hits in both repos — four surfaces, one wording. Worth noting I'd assumed the android side would narrow to match iOS; converging on the long form is the better result.
• Still different, and correctly so: this repo keeps "Missing or uncertain" where android drops "missing". Android's screens map makes a missing frame unreachable; you have no map, so "missing" is a live case here. Not asking you to match.

I also re-checked that the pr.md:115 gate doesn't contradict the unconditional "must … under ### Design" at AGENTS.md:342-344 — it doesn't, since AGENTS.md is repo-local and this template carries the heading. Android has the identical two-layer shape.

Clean from my side.