release: beta.9 notes misdescribe the #5985 security fix

[taliesin-ai]

Evidence

v3.0.0-beta.9 was published in the 2026-08-16 release window. Its Fixed section says:

Fixes issue with incorrect handling of empty strings in JSON parsing in PR #5985

PR #5985 was actually fix(security): update website nanoid lockfiles, resolving the two high-severity website-toolchain alerts tracked by #5983. It contains no JSON empty-string parsing change.

The incorrect text originated in auto-changelog commit da13deffde44bcd4efc0fad1f9b8ae377e2de499 and was folded into the beta.9 changelog/release notes. Duplicate issue/PR searches for beta.9, #5985, nanoid, and the incorrect text found no existing tracker.

Impact

• The public beta.9 notes misdescribe a security dependency remediation.
• The committed v3 changelog preserves the same false statement.
• Readers cannot tell from the release notes that the nanoid alerts were remediated.

Scope

Correct the committed changelog source through a focused PR. Do not mutate the published release, tag, or release workflow from automation; the published beta.9 body needs a maintainer-owned correction. Add a narrow regression guard if the auto-changelog path can verify that generated text is consistent with the source PR without introducing another heuristic failure mode.

Acceptance criteria

• The committed beta.9 changelog describes fix(security): update website nanoid lockfiles #5985 as the website nanoid lockfile security update.
• The corrected entry links fix(security): update website nanoid lockfiles #5985 and credits the author accurately.
• A maintainer corrects the published beta.9 release body, or explicitly records why it will remain immutable.
• Any automated guard is covered by focused tests and rejects this exact title/body mismatch.
• No gate is passed or waived from this correction alone.

[taliesin-ai]

A focused correction and regression guard are pushed in PR #5993 (current head 14d688b09). The PR restores the beta.9 entry as the website nanoid 3.3.18 security update, rejects PR #5985's status-only CodeRabbit comment as changelog context, always includes the source PR title, and lets the historical validator accept only same-PR bullet replacements while continuing to reject new, source-less, and different-source released entries.

Evidence: both focused Go test commands pass locally; an integration run against the exact beta.9 added/deleted lines returned VALIDATION_RESULT=success without changing the file; git diff --check passes; and the updated GitHub Changelog Validation run passed without moving the correction.

CodeQL then reported a real high alert on the initial unanchored PR URL regex (check run 95314377380): an attacker-controlled URL could contain the trusted GitHub URL as a substring. Commit 14d688b09 fixes this by extracting only the [PR](...) destination and requiring the entire destination to match the canonical Wails PR URL; focused tests reject hostile host prefixes, suffixes, and query strings. Local focused tests and the exact-diff integration validation pass after the fix. Hosted checks for the new head are running.

The required coderabbit --plain invocation was attempted but this installed CLI does not support the flag; its supported review fallback hit the free OSS review limit. The published beta.9 release body and tag were not mutated and still require a maintainer decision.

[taliesin-ai]

Execution evidence (2026-08-19): CodeRabbit identified a critical workflow-dispatch credential boundary in PR #5993, a one-sided-diff failure, and then a cross-job output injection edge. Commits cfc49e3 and d19feb8 remove repository-write access and persisted credentials from PR-controlled validation, move manual reporting to a separate no-checkout job, remove automatic pushes, make both diff filters no-match-safe, and pass report values as environment data rather than JavaScript source. Focused auto-changelog and validator tests, YAML parsing, privilege/source-injection assertions, additions-only/deletions-only cases, the real #5993 correction, and git diff checks pass. coderabbit --plain was run before both commits but this installed CLI rejects the obsolete flag; the first supported local review returned no findings and the second was rate-limited after hosted CodeRabbit found the final edge. actionlint and zizmor are unavailable locally. All three new review threads are resolved. The PR remains In Progress pending fresh hosted CI and independent review.