docs: CHANGELOGs + freshness riders conformance wave (10 plugins)

[kyle-sexton]

Summary

Wave #316 of epic #313 (conformance matrix dims 7 + 14).

CHANGELOGs (dim 7) — the last 3 plugins without one now ship a Keep-a-Changelog file seeded with the current version entry: firecrawl 0.2.2, miro 0.1.2, playwright 0.2.1. Every versioned plugin in the fleet now has a CHANGELOG. Miro's version drift is resolved on this bump: package.json, the MCP server Implementation version, the lockfile, and the rebuilt committed bundle all report 0.1.2 (bundle drift check, typecheck, and the 32-test vitest suite green; lockfile delta is the two version fields only).

Freshness riders (dim 14) — all 7 flagged plugins now carry verified-date + official-link riders on their volatile platform facts. Every fact was re-verified this session before stamping; where verification failed, the fact was corrected first (doctrine: never date a stale claim):

• context7 0.3.1 — CLI facts corrected against live ctx7 0.5.5: --base-url default is https://context7.com (not /api), version flag is lowercase -v, library's query is optional, the upstream skills surface is deprecated (this plugin never invokes it), new remove/uninstall + upgrade commands listed. The Claude Code unset-env MCP claim was rewritten per current docs: the config loads with a missing-variable warning and sends the literal ${VAR} text — silently broken auth, not a parse failure.
• songwriting 0.4.1 — the May lyrics-limit consensus flipped: current third-party testers agree on a 5,000-char hard cap on v4.5/v5/v5.5 (3,000 was the v4-era cap); ~3,000 stays as the documented quality sweet spot. Title corrected to ~100 chars; tier-matrix Studio row corrected to Premier-exclusive (fixes an internal contradiction with studio.md). Unverifiable rows keep their hedges undated; the conflicting upload-length figures are cross-flagged, not resolved. All help.suno.com source links fixed to the working /en/articles/ form (all 14 verified 200). Riders state explicitly that char limits are third-party-tester sourced — Suno publishes no official field limits.
• claude-ops 0.11.3 — scope-semantics doc gains a dated verification header; all version gates re-verified (/reload-plugins --force ≥ 2.1.163, renames ≥ 2.1.193, plugin prune ≥ 2.1.121, userConfig type set per the published schema). The monitor-restart claim now cites its actual source (plugins-reference). Setup's pluginConfigs claim dated + pinned to ≥ 2.1.207.
• desktop-notification 0.3.2 — terminalSequence v2.1.141+ claim verified against the Claude Code changelog, dated, linked. Also backfills the missing 0.3.1 CHANGELOG entry (feat(guardrails): add block-dangerous-git guard (0.6.0) #298's hook-utils resync bumped the version without one).
• skill-quality 0.4.1 / verification 0.2.1 — pluginConfigs (≥ 2.1.207) and bundled /verify + /run (≥ 2.1.145, per the official bundled-skills min-version marker) claims verified, dated, linked.
• ai-briefing 0.5.2 — README Playwright matrix confirmed current and dated against the official system-requirements page; the setup skill no longer restates the volatile matrix and defers to the linked page (evidence rule 5).

Verification

• validate-plugins.sh, validate-plugin-contracts.mjs (1473 files), markdownlint (176 files): green.
• All external links added/touched return 200; the one bot-blocked npmjs link was dropped in favor of the official CLI docs link.
• All test suites in touched plugins pass. Note: telemetry/slow-sink timing assertions (desktop-notification, markdown-format) flake locally on Windows under process-spawn load (1.6–9.4s variance vs 2s threshold) — unrelated to this wave (markdown-format is untouched) and consistently green on CI.

Related

• Closes Conformance wave: CHANGELOGs + freshness riders (15 + 7 plugins) #316
• Part of Epic: plugin fleet conformance audit — 50-plugin doctrine matrix and remediation waves #313

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4664b072fa

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 3m 20s —— View job

---

PR Review

• Gather context and understand the request
• Read changed files and analyze the diff
• Review CHANGELOG additions (firecrawl, miro, playwright)
• Review freshness riders for correctness and completeness
• Review miro version drift fix
• Check for security, correctness, and alignment with CLAUDE.md
• Post final review findings

---

Summary

This is a clean, well-scoped documentation conformance wave. All 10 plugins get correct minor version bumps with matching CHANGELOG entries. No security concerns. The issues below are all documentation-level and none are blocking.

---

Miro version drift fix

The version alignment is complete and internally consistent:

Artifact	Before	After
plugin.json	0.1.1	0.1.2
package.json	0.1.0	0.1.2
src/index.ts MCP Implementation	0.1.0	0.1.2
dist/index.min.js	(0.1.0 bundle)	(0.1.2 bundle)
package-lock.json	updated	✓

package.json jumping from 0.1.0 directly to 0.1.2 (skipping 0.1.1) is intentional and correctly explained in CHANGELOG.md: 0.1.1 was a plugin.json-only bump; the npm package was never published at 0.1.1.

---

Context7 CLI corrections

Material correctness fixes verified against live ctx7 0.5.5:

• cli.md: --base-url default corrected (https://context7.com/api → https://context7.com) and version flag corrected (-V → -v). Both are observable failures in the old doc.
• mcp.md: The Claude Code unset-env-var behavior correction is significant — "fails to parse the config" was wrong; the actual behavior (loads with a warning, sends literal ${VAR} text, silently broken auth) is a much more dangerous failure mode that users need to understand correctly.

---

Songwriting / Suno

• Lyrics cap flip is well-sourced and the confidence downgrade (HIGH → MEDIUM-HIGH + "third-party tester consensus only") is exactly right since Suno publishes no official field limits. The historical trace of the position flip in SKILL.md is genuinely useful for future readers evaluating source credibility.
• Studio tier fix: correcting partial/full/full → —/—/✓ (Premier-exclusive) eliminates an internal contradiction with studio.md that would have confused the model.
• /en/articles/ URL fixes: All 14 help.suno.com source links corrected. Confirmed 200 per the PR.

One open observation on the audio upload row — consistent with the Codex review comment:

v55-features.md line 99 still has 60s | 120s | 120s. Line 110 already cross-flags this as conflicting with an 8-minute figure cited elsewhere in the skill. The Codex reviewer notes the current Suno pricing page apparently shows 8 min (Free) and 30 min (Pro/Premier), which would make the table even more stale. The PR deliberately defers resolution per the "never date a stale claim" doctrine — the cross-flag note is correct and visible — but this whole row deserves a dedicated follow-up verification pass against the live pricing page before anyone relies on it. Consider a TODO comment or linking the Codex comment to a follow-up issue.

---

Freshness riders

The rider format is consistent across all 7 plugins: Verified YYYY-MM-DD against <url> with a re-check reminder. The ai-briefing setup skill correctly replaces a hardcoded volatile matrix with a live link to Playwright's system requirements (evidence rule 5 application). The pluginConfigs ≥ 2.1.207 and bundled /verify + /run ≥ 2.1.145 version pins are now sourced and dated — these were the right facts to lock down.

---

CHANGELOG format

All new and updated CHANGELOGs follow the established fleet convention (version headers without dates, matching the pattern in every existing CHANGELOG). Keep-a-Changelog recommends dates but the fleet has a consistent dateless style — this PR is coherent with it.

---

Minor nits (non-blocking)

1. SKILL.md:171 — the instruction Stay ≤3,000 chars — the quality sweet spot; the hard cap is 5,000 — unless user asked for max is grammatically awkward with two em-dashes in one clause. Suggested: Stay ≤3,000 chars (quality sweet spot; hard cap is 5,000) unless user asked for max. Fix this →

2. v55-features.md:111 — Free-tier generation runs on **v4.5-all**, not v5.5 (third-party report: TechRadar) is a new unverified claim inserted without a source link. The TechRadar article should be hyperlinked if kept, or the rider should explicitly say "unverified" like adjacent claims do. Fix this →

---

Verdict: Approve with the two nits above as suggestions. The audio upload row follow-up is the only item worth tracking as a future issue; it's already documented in-file and is not a regression introduced by this PR.