v1.9.8: Manifest Consistency Guard + Uninstall Fix

v1.9.8: Manifest Consistency Guard + Post-v1.9.7 Polish

A same-day patch release on top of v1.9.7 that fixes a recurring skill-count drift, locks in two post-tag polish commits that were intended for v1.9.7 but landed too late, and adds a CI guard so this class of drift cannot ship again.

Why a same-day v1.9.8

1. Skill-count drift returned via PR #56. The reconciliation in v1.9.7 locked the canonical phrasing at "20 core" sub-skills. After Phase A locked it, PR #56 (seo-content-brief by @puneetindersingh) merged a 21st core skill, but the canonical phrasing was not re-run. v1.9.7 shipped with manifests + docs claiming 24 sub-skills when reality is 25.
2. Two commits landed on main after the v1.9.7 tag. 8514999 (marketplace metadata polish, added category, author.email, docs-site homepage, 14-keyword array) and 66a7485 (em-dash sweep on user-visible AGENTS.md and CHANGELOG.md) were scoped at v1.9.7 but landed post-tag. v1.9.8 captures them in a tagged release.

Fixed

• Skill-count reconciliation. plugin.json, marketplace.json (both descriptions), CLAUDE.md, AGENTS.md, README.md, and docs/ARCHITECTURE.md now all reference 25 sub-skills (21 core + 1 orchestrator + 1 framework integration + 2 extension mirrors). Math: 21 + 1 + 1 + 2 = 25. Verified against ls -d skills/*/.

Added

• tests/test_manifest_consistency.py with 7 assertions:
  1. plugin.json description's N sub-skills claim equals on-disk skill directory count
  2. plugin.json description's N sub-agents claim equals on-disk agent file count
  3. marketplace.json plugin entry's sub-skills count matches plugin.json
  4. marketplace.json plugin entry's sub-agents count matches plugin.json
  5. README, CLAUDE.md, AGENTS.md all reference the canonical sub-skills count
  6. plugin.json version equals CITATION.cff version
  7. Canonical math adds up (parenthetical breakdown sums to the headline number)
• pytest tests/ job in .github/workflows/ci.yml runs the new suite plus the existing test_sync_flow.py on every push to main and every pull request. Uses GH_TOKEN to avoid anonymous GitHub API rate limiting.

Changed

• uninstall.sh and uninstall.ps1 now use glob enumeration rather than a hardcoded skill list. The previous scripts had been frozen at v1.4.0-era state and missed 12 sub-skills and 11 sub-agents added between v1.5 and v1.9.8. Anyone running the old uninstaller got half a cleanup. Glob enumeration (skills/seo, skills/seo-*, agents/seo-*.md) auto-tracks future additions without requiring uninstaller maintenance per release. Sandbox-tested with 8 assertions: removes every seo and seo-* skill plus every seo-*.md agent while leaving sibling skills (e.g. blog-writer, security) untouched.

This release also rolls forward two commits that landed on main after the v1.9.7 tag:

• 8514999 marketplace metadata polish: added category: "marketing", author.email, homepage: https://claude-seo.md, and a 14-keyword array on the marketplace.json plugin entry. These are read by Anthropic's marketplace curation and improve directory discoverability.
• 66a7485 em-dash sweep on user-visible AGENTS.md and CHANGELOG.md. Skills, agents, and extension internals deliberately preserved (those are LLM-readable instruction text and careless replacement could subtly change behavior).

Migration

None required. Drop-in patch release.

Dependencies

No runtime dependency changes. The five Dependabot bump PRs queued in the v1.9.7 release window remain queued for a v1.9.9 follow-up batch.

Test count

22 tests pass: 7 new manifest-consistency tests plus 15 existing in test_sync_flow.py.

Full changelog

v1.9.7...v1.9.8