Release status: v2.2.3 was superseded by Claude SEO v2.2.4 on July 20, 2026. This page remains the historical record for the v2.2.1 to v2.2.3 maintenance train.
Claude SEO v2.2.3 completed a three-release maintenance train. It continues the v2.2.0 security and reliability work through v2.2.1, v2.2.2, and v2.2.3. It refreshes time-sensitive Google guidance, corrects command and extension wiring, adds a strict consistency gate, and finishes with prompt hygiene. The release notes are explicit: no breaking changes
.
- Current guidance: Lighthouse, WebMCP, AI search, Search Console, schema, crawling, international SEO, and local SEO references were refreshed.
- Correct commands: routing, flags, tool names, shared references, and extension documentation were audited across 25 skills and 8 extensions.
- Enforced consistency: a new gate checks references, routing tables, FLOW integrity, versions, portability, and orphaned resources.
- Prompt hygiene: v2.2.3 removes unnecessary CAPS emphasis, replaces dash punctuation in prose, and trims five descriptions without changing behavior.
- Verified release: 328 tests passed, plus manifest, version, portability, syntax, secret, and private-reference checks.
What changed across v2.2.1 to v2.2.3?
The versioned changelog shows one clear sequence. First, bring volatile SEO guidance and command documentation back to the source of truth. Second, turn those consistency expectations into checks. Third, clean the prompt surface without altering routing or outputs.
v2.2.1 refreshes Google guidance and command truth
v2.2.1 updates Lighthouse to 13.4.0, documents the Agentic Browsing score as a fractional pass ratio, and replaces future WebMCP wording with the Chrome 149 origin trial and its three Lighthouse audits. GEO guidance now reflects AI Overviews reach, Google's treatment of llms.txt, and the fact that AI search products use distinct citation engines.
The same pass updates Search Console Generative AI reporting, GA4 AI Assistants, Core Web Vitals and technical SEO, schema status, sitemap guidance, Googlebot behavior, international SEO, Google Business Profile, and Ask Maps. It also corrects command flags, MCP tool names, Bing host flags, shared paths, and documentation across the core skill set and eight extensions.
v2.2.2 makes correctness testable
v2.2.2 applies 319 verified findings from seven research streams, 20 currency slices, and six refuter slices. Concrete corrections include removing a false Google Business Profile Q&A deprecation, describing AI Mode as a custom version of Gemini 2.5, refreshing Gemini image model IDs, repairing Banana script references, and fixing WebMCP audit identifiers used by the parallel audit workflow.
The durable change is scripts/consistency_check.py. It checks dead references, routing tables, FLOW integrity, and orphaned resources. The schema hook stdin contract was also fixed, so the validation layer now checks both documentation consistency and an actual integration boundary.
v2.2.3 cleans the prompt surface
v2.2.3 removes CAPS emphasis words while preserving severity tokens, replaces em and en dash prose outside code fences, and trims five overlong skill descriptions. According to the changelog, it changes no behavior, routing, or output. This is a small diff with a useful goal: instructions should be direct without adding noise that can distort model attention.
How was the release validated?
The most important v2.2.3 improvement is not another command. It is evidence that the existing surface stays synchronized. The release reports a clean test suite, exact version alignment, strict consistency, portable skills, valid Python syntax, and a clean secret scan.
| Release gate | Result |
|---|---|
| Full test suite | 328 passed |
| Manifest and version consistency | 15 passed |
| Strict consistency | 0 errors and 0 warnings across 364 tracked files |
| Strict portability | 0 findings across 33 SKILL.md files |
| Python syntax | All 51 tracked scripts passed |
| Secret scan | Clean |
Release integrity check, verified July 20, 2026: public main and the dereferenced annotated v2.2.3 tag both resolve to commit 9b8eb69. The annotated tag object has its own SHA, which is normal Git behavior. The public tagged tree does not contain outputs/ or tmp/.
How do you install v2.2.3?
For a fresh Claude Code plugin install, add the marketplace and install Claude SEO:
/plugin marketplace add AgriciDaniel/claude-seo /plugin install claude-seo@agricidaniel-claude-seo
For a manual fresh install, clone the current public repository and run the installer:
git clone --depth 1 https://github.com/AgriciDaniel/claude-seo.git bash claude-seo/install.sh
The official repository installation guide documents the current upgrade flow, and the short installation page keeps the supported platform commands together. Review downloaded scripts before running them, especially when a convenience command pipes remote content directly into a shell.
Why this maintenance release matters
SEO instructions age quickly because Google documentation, browser audits, model names, reporting surfaces, and schema support all move on different schedules. A skill can keep producing polished reports while its citations, flags, or routing quietly drift. v2.2.3 attacks that failure mode twice: researched corrections update the knowledge layer, and automated checks guard the connections between skills, commands, extensions, and references.
That makes this release more useful than its patch number suggests. Teams get current guidance without retraining their workflow, maintainers get failures when metadata or references diverge, and users can trace the public release back to the exact commit that is on main.
Official sources
- Claude SEO v2.2.3 release, retrieved July 20, 2026.
- CHANGELOG at v2.2.3, retrieved July 20, 2026.
- Release commit 9b8eb69, retrieved July 20, 2026.
- GitHub Actions release validation, retrieved July 20, 2026.
- Installation guide at v2.2.3, retrieved July 20, 2026.