[Atom-oh]

This prerequisite changes the MDX truncation marker in 60 existing release posts across Korean, English, Japanese and Chinese to the JSX form supported by both Docusaurus 3.9.2 and the refresh target. Published text is preserved.

All 60 posts retain the exact bytes from #224's original integrated snapshot 63cf803083d06a5d90289fb9238b9fb55d5b0abe. Separating these marker changes allows both PRs to receive complete native review under the existing 128 KiB limit. Application dependencies, runtime changes, documentation toolchain/configuration and CI workflows are now tracked by the ordered replacement stack #227 → #228 → #229 → #230; the author closed #224 after splitting it. No review limit or check changes.

The initial prerequisite also copied an initiative record into docs/superpowers, which this repository reserves for private-upstream citations. That file has been removed; no completed runtime-refresh checklist is published by this documentation-only prerequisite.

Validation of the marker source using the existing dev lockfile:

• Documentation TypeScript check and all four locale production builds passed.
• Exactly 60 one-line marker replacements, 15 per locale; no prose or documentation configuration changes.
• Documentation path checks passed after removal of the private-path record. No tracked inbound references remain.

Integration waits for a fresh complete native dual-family review, current CI and an independent publication content gate. A merge to dev triggers the existing Pages workflow. The replacement stack will incorporate the reviewed dev changes and receive full native review for each stage after retargeting to dev.

[github-actions]

🤖 AI Code Review (two independent reviewers)

_Reviewer responses: codex/ALL claude/ALL _

Status: PASSED — No blocking issues found

All local evidence is verified. I have what I need to synthesize the chair review.

Chair Review — PR #225: docs: prepare MDX release notes for dependency refresh

Summary

This PR changes exactly one line in each of 60 release-note files under docs-site/ (15 Korean sources in whatsnew/ plus their 15-file en/ja/zh translations), replacing the HTML-comment excerpt marker <!--truncate--> with the MDX expression-comment form {/* truncate */}. I verified against the base tree that it covers all 60 occurrences of the old marker with no stragglers and no over-reach (the two <!-- strings remaining in docs-site/whatsnew/2026-03-29-v1.7.1.md:121,128 are inside fenced code samples and correctly untouched). The one substantive disagreement between the panels — whether Docusaurus still recognizes the new marker — resolves against the codex finding.

Issues per lens

L2 — Code correctness

• Codex flagged MAJOR / Claude explicitly refuted — chair rejects the MAJOR as a false positive. Codex claimed the default truncateMarker "matches HTML comments, not JSX comments," which would break excerpt truncation on the /whatsnew indexes and feeds. That describes Docusaurus v2. This site pins @docusaurus/preset-classic 3.9.2 (docs-site/package.json) and does not override truncateMarker (docs-site/docusaurus.config.ts:62-71, verified — the blog preset options set only path/title/sidebar/readingTime). In Docusaurus 3.x the default marker regex in @docusaurus/plugin-content-blog accepts both forms (<!--\s*truncate\s*--> or \{\/\*\s*truncate\s*\*\/\}); the MDX form was added in v3.0 precisely because .md files are MDX-compiled by default, it is the form the official v3 blog docs and create-docusaurus scaffold templates use, and the new lines match the \s* groups exactly. Caveat for the record: node_modules is not installed in this worktree and the sandbox blocks network fetches, so I could not grep the installed package source directly; the conclusion rests on the version-pinned config plus well-documented 3.x defaults, and Claude's independent version-specific analysis agrees. Codex cited no source and offered no 3.x-specific evidence. Excluded from the gate.
• MINOR (one model, Claude; chair concurs, non-blocking): truncation has no regression guard — if the marker ever stops matching, npm run build (run by merge-verify.yml:42-47 on docs changes) still succeeds and the only symptom is full posts on the blog index/feeds. A one-line assertion on the built whatsnew/index.html would make this fail-fast. Suggestion only.

L3 — Security / AWS mutation safety

No findings — both models agree. No AWS calls, IAM, Terraform, secrets, or runtime code are touched; the ADR-005 mutation freeze is not implicated. The changed lines contain only the literal marker text, and the SRI-pinned external stylesheet in docusaurus.config.ts is unmodified. If anything, the MDX comment form is stripped at compile time where the HTML comment shipped as a comment node in output HTML — a marginal reduction in incidental output.

L4 — Observability / data-integration correctness

No findings — both models agree. No Steampipe/Powerpipe SQL, CIS checks, AgentCore routing/diagnosis logic, or datasource query generation is modified, and none of the quantitative claims in the release-note prose change.

L5 — Docs/ADR consistency

• No findings — both models agree. All four locales migrated together in the same PR (15 × 4 = 60, verified complete against base), satisfying the CLAUDE.md rule that multilingual docs-site/ content retains locale parity. Front matter, headings, and :::tip blocks are untouched, so tags/authors/sidebar/feed metadata are unaffected. No repo authoring doc prescribes the old marker, so no companion doc edit is missing. CHANGELOG.md is not touched, which is correct — this is not net user-visible product behavior.
• MINOR (pre-existing, out of diff scope, excluded from gate): only i18n/zh has a blog options.json (verified: en/ja lack it), so Korean blog title strings leak into those locales. Predates this PR; noted only to avoid mistaking it for a regression.

Suggestions

1. Add a cheap truncation canary to CI (assert the built /whatsnew index still contains a "Read More"/truncated excerpt) so future marker or markdown.format changes fail fast instead of silently rendering full posts.
2. Follow-up (separate PR): add options.json blog translations for en and ja to match zh.

Verdict

The only CRITICAL/MAJOR candidate (codex's truncate-marker regression) is contradicted by the pinned Docusaurus 3.9.2 configuration and 3.x default behavior; no other blocking finding exists. The image manifest is empty with status complete and no unavailable/omitted entries.

IMAGE_COVERAGE: NOT_REQUIRED

---

Triggered by commit 3a596bdc61b7a4903d51bb6d359f3c1e0f7d0a2d · workflow: .github/workflows/pr-review.yml