valhalla.md

[3.0.0] - 2026-07-22

Valhalla release. Adds a single-command front door and renames the plugin identity. Full suite: 480 passing tests.

Added

• Valhalla — a root SKILL.md (name: valhalla) that is the one-command entry point. bmo add justin06lee/claude-seo.bmo (bare, no subpath) now installs a single valhalla skill: the whole toolkit ships as knowledge Valhalla drives internally, so the user gets exactly one command, /valhalla, which runs the full autonomous SEO pass over a codebase. A .bmoignore keeps that single-skill install lean.

Changed

• BREAKING: plugin identity renamed to valhalla. plugin.json name and the marketplace slug are now valhalla / justin06lee-valhalla; plugin install is /plugin install valhalla@justin06lee-valhalla. The discipline skills are unchanged and, via a plugin or …/skills --all install, still register individually (/seo audit, etc.).
• README leads with the Valhalla one-command install; the granular skills are documented as the advanced path.

Notes

• Two install shapes coexist from one tree: bare bmo add → single valhalla skill; plugin or …/skills --all → the 26 discipline skills registered individually. The plugin loader ignores the root SKILL.md because a skills/ directory is present.

[2.5.0] - 2026-07-22

Feature release: an autonomous full-project SEO autopilot. Full suite: 480 passing tests.

Added

• /seo everything [path] — a new seo-everything skill that takes a codebase from wherever it is to as SEO-complete as the project allows, with no per-skill decisions from the user. It perceives the project with a deterministic scanner (seo_inventory.py: framework + router, routes, existing SEO surface, business type), decides which disciplines apply, fans out read-only specialists in parallel, then applies every relevant fix to the source on a feat/seo-pass branch — idempotently, idiomatically for the framework, without breaking the build — and leaves one diff to review. Three references cover per-framework SEO placement, the signal-to-discipline relevance matrix, and apply safety (git isolation, never fabricate a fact, never touch secrets or build output, honor the Google-currency guardrails).
• seo_inventory.py: a stdlib-only, read-only codebase SEO scanner, covered by 16 tests.

[2.4.0] - 2026-07-22

Feature release: scheduled/CI drift monitoring, a cross-site portfolio dashboard, and a renderer that hydrates scroll- and interaction-gated content. Full suite: 464 passing tests.

Added

• Scheduled / CI drift runner (drift_ci.py). Watches a list of URLs non-interactively, seeds baselines on first sight, compares afterward, and exits non-zero when a severity threshold is breached, so cron or CI fails on SEO regression like a broken test. JUnit output for CI display; CLAUDE_SEO_DRIFT_DIR gives reproducible baselines across runners. New skill command /seo drift ci and a references/ci-integration.md workflow guide.
• Cross-site portfolio dashboard (portfolio_report.py). Aggregates the audit-data.json each audit writes under {domain}-audit/ into one standalone HTML view: sites ranked worst-first by health score, findings totaled by severity, issues that recur across sites surfaced as fix-once candidates, and per-site score deltas against a previous snapshot.

Changed

• Renderer hydration. render_page.py gains a settle strategy applied after navigation: scroll (the new default) walks the page to fire IntersectionObserver and scroll-bound hydration, network adds a bounded network-idle wait for race-prone widget mounts, and opt-in --reveal-hidden expands <details> and aria-expanded tab/accordion controls without following links or submitting forms. Closes the documented false negatives for below-the-fold and interaction-gated content. Result adds settle_strategy, scroll_passes, and revealed_elements; verified against real headless Chromium.

[2.3.0] - 2026-07-21

Fork release. This repository is a distribution of AgriciDaniel/claude-seo; upstream authorship, the MIT license, and contributor credit are unchanged. Full suite: 415 passing tests.

Changed

• The seo skill now carries its own runtime. scripts/, agents/, schema/, pdf/, data/, bin/, and requirements.txt moved into skills/seo/, which is the directory the runtime already treated as its root. Installing that one folder now yields a working whole, which is what makes skill-granularity installation possible.
• .claude-plugin/plugin.json declares all 18 subagents explicitly. Claude Code's agents key takes file paths and replaces the default agents/ scan, so a subagent that is not listed does not exist for plugin users; a test pins the list to disk.
• bin/claude-seo at the repository root is now a shim that execs the launcher inside the skill. Claude Code adds only <plugin-root>/bin to PATH and offers no manifest field to relocate it, so the shim is what keeps the command working for plugin installs.
• Distribution pointers (repository URLs, marketplace slug, installer tag) name this fork. hooks/ stays at the plugin root, where it is auto-discovered.
• CLAUDE.md documents the fork's actual single-remote topology. The previous public/private two-remote workflow described a setup this repository does not have, and docs/WORKFLOW-public-private.md was removed with it.

Added

• Installation via bmo: bmo add justin06lee/claude-seo.bmo/skills --all installs all 25 skills and their 18 subagents in one command.

Fixed

• The runtime resolves its version in a skill-folder install, where no plugin manifest is present, by reading the enclosing plugin root when nested in one and otherwise the version the skill declares in its own frontmatter.