feat(skills+mcp): humanizer skill mirror + mcp-servers ref expansion (v6.50.0)

CHANGELOG.md

@@ -7,6 +7,67 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [6.50.0] - 2026-05-20
+
+Two small additions: `humanizer` skill from `blader/humanizer` (Wikipedia
+"Signs of AI writing" cleanup, 20k★ upstream) joins the skills marketplace
+mirror; `components/mcp-servers-guide.md` upgrades its bare
+`awesome-mcp-servers` reference into a proper discovery-surface explanation.
+
+### Added
+
+- **`templates/skills-marketplace/humanizer/SKILL.md`** — mirror of
+  [blader/humanizer](https://github.com/blader/humanizer) at upstream commit
+  `8b3a17889fbf12bedae20974a3c9f9de746ed754`. v2.5.1 skill that detects and
+  removes AI-writing patterns (inflated symbolism, promotional language,
+  superficial -ing analyses, vague attributions, em-dash overuse, rule of
+  three, AI vocabulary words, passive voice, negative parallelisms, filler
+  phrases). Based on Wikipedia's WikiProject AI Cleanup guide. MIT license.
+- **`manifest.json:skills_pins.humanizer`** — drift-tracking pin with
+  sha256 `6e2bb628772b87750849d74e1176736c405f399ab4ebbf3f08e1ce26ed69852c`.
+  Total active pins: 23 (was 22) + 1 no-upstream-found.
+- **`files.skills_marketplace`** registration so the skill installs via
+  `init-claude.sh` / `update-claude.sh`.
+
+### Changed
+
+- **`components/mcp-servers-guide.md`** — the trailing
+  `awesome-mcp-servers` link was a bare URL. Replaced with a one-sentence
+  explanation of what the upstream index covers (databases, browser
+  automation, finance, dev tooling, knowledge bases, communication), why
+  the toolkit does NOT vendor the 24 MB index, and the install line to use
+  once a server is picked. No link target change.
+
+### Why
+
+- `humanizer` covers a gap the existing `ru-text` / `copy-editing` skills
+  don't: detection of AI-pattern English prose by structural heuristics
+  rather than language-specific style. Useful on AI-drafted READMEs, copy,
+  and commit prose where a human reviewer wants the "did an LLM write
+  this?" signal compressed into actionable edits.
+- `awesome-mcp-servers` is the canonical community discovery index. The
+  toolkit's `Recommended Servers` section is curated for the toolkit's
+  typical workloads; the awesome-list covers the rest. The bare URL hid
+  this from readers who didn't know the upstream existed.
+
+### Files changed
+
+- New: `templates/skills-marketplace/humanizer/SKILL.md`
+- Updated: `manifest.json` — version `6.49.0` → `6.50.0`, new
+  `skills_pins.humanizer`, `files.skills_marketplace` adds humanizer,
+  `skills_pins_note` count `22 → 23 active`.
+- Updated: 9 installer scripts — `TK_TOOLKIT_REF` default bumped.
+- Updated: `components/mcp-servers-guide.md` — awesome-mcp-servers
+  reference expanded from bare URL into a discovery-surface explanation.
+
+### Upgrade path
+
+- End users: `/update-toolkit` pulls the humanizer skill + the refreshed
+  mcp-servers guide. The skill installs under
+  `~/.claude/skills/humanizer/SKILL.md` and registers in the skills
+  discovery surface. Invoke it as a skill on any text the user wants
+  cleaned of AI patterns.
+
 ## [6.49.0] - 2026-05-20
 
 New integration: `awesome-design-md` brand catalog mirrored into the toolkit

components/mcp-servers-guide.md

@@ -515,6 +515,6 @@ rtk discover          # Analyze Claude Code history for missed opportunities
 
 ## Useful Resources
 
-- [MCP Specification](https://modelcontextprotocol.io/)
-- [Claude Code MCP Docs](https://docs.anthropic.com/en/docs/claude-code/mcp)
-- [Awesome MCP Servers](https://github.com/punkpeye/awesome-mcp-servers)
+- [MCP Specification](https://modelcontextprotocol.io/) — canonical protocol spec.
+- [Claude Code MCP Docs](https://docs.anthropic.com/en/docs/claude-code/mcp) — Claude Code's own MCP integration guide.
+- [`punkpeye/awesome-mcp-servers`](https://github.com/punkpeye/awesome-mcp-servers) — community-curated index of public MCP servers across categories (databases, browser automation, finance, dev tooling, knowledge bases, communication). Use it as a discovery surface when the toolkit's `Recommended Servers` section above does not cover your need. The list is NOT vendored into the toolkit (the awesome-list is large, ~24 MB, and updates daily); follow links upstream, then `claude mcp add -s user <name> -- <command>` once you pick a server.

manifest.json

@@ -1,6 +1,6 @@
 {
   "manifest_version": 2,
-  "version": "6.49.0",
+  "version": "6.50.0",
   "updated": "2026-05-20",
   "build_date": "2026-05-20",
   "description": "Claude Code Toolkit manifest for smart updates",
@@ -63,7 +63,7 @@
       "mirror_subset": "design-md/<brand>/DESIGN.md (71 brands; README.md per brand intentionally not mirrored)"
     }
   },
-  "skills_pins_note": "Pinned upstream HEAD of mirrored marketplace skills. Used by scripts/update-deps.sh probe_skill_pin to detect drift between this mirror and the canonical upstream repo. Current state (v6.46.0, 2026-05-18): 22 active pins + 1 no-upstream-found (memo-skill, exhaustive gh search yielded no canonical upstream \u2014 possibly toolkit-original or upstream deleted). Schema fields: repo, path (optional, for monorepo subpath probes via GitHub API), tag, commit, pinned_at, sha256 (mirror content checksum, reproducible across BSD/GNU via scripts/lib/skill-checksum.sh), _status \u2208 {active, no-upstream-found}. Trajectory: v6.35.0 (2 pins) \u2192 v6.37.0 (+path field, +3 anthropics monorepo) \u2192 v6.41.0 (+4 vercel/firecrawl) \u2192 v6.44.0 (+12 confirmed via gh code search + majiayu000/claude-skill-registry cross-reference) \u2192 v6.46.0 (+sha256 field, closed-loop sync, mirror\u2194manifest drift detection) \u2192 v6.47.0 (hardened sync: tab/newline-safe checksum, atomic mirror swap, post-checkout SHA verification, 3-way CLEAN/SOFT/DRIFT classification via --normalize so markdownlint-style cosmetic diffs do not trip --strict CI gate). Add new entries one at a time as upstream URLs are confirmed; do not invent. validate-manifest.py enforces note\u2194data consistency: any count referenced here must match the live skills_pins map AND every mirror sha256 must match the manifest-declared sha256.",
+  "skills_pins_note": "Pinned upstream HEAD of mirrored marketplace skills. Used by scripts/update-deps.sh probe_skill_pin to detect drift between this mirror and the canonical upstream repo. Current state (v6.50.0, 2026-05-20): 23 active pins + 1 no-upstream-found (memo-skill, exhaustive gh search yielded no canonical upstream \u2014 possibly toolkit-original or upstream deleted). Schema fields: repo, path (optional, for monorepo subpath probes via GitHub API), tag, commit, pinned_at, sha256 (mirror content checksum, reproducible across BSD/GNU via scripts/lib/skill-checksum.sh), _status \u2208 {active, no-upstream-found}. Trajectory: v6.35.0 (2 pins) \u2192 v6.37.0 (+path field, +3 anthropics monorepo) \u2192 v6.41.0 (+4 vercel/firecrawl) \u2192 v6.44.0 (+12 confirmed via gh code search + majiayu000/claude-skill-registry cross-reference) \u2192 v6.46.0 (+sha256 field, closed-loop sync, mirror\u2194manifest drift detection) \u2192 v6.47.0 (hardened sync: tab/newline-safe checksum, atomic mirror swap, post-checkout SHA verification, 3-way CLEAN/SOFT/DRIFT classification via --normalize so markdownlint-style cosmetic diffs do not trip --strict CI gate). Add new entries one at a time as upstream URLs are confirmed; do not invent. validate-manifest.py enforces note\u2194data consistency: any count referenced here must match the live skills_pins map AND every mirror sha256 must match the manifest-declared sha256.",
   "skills_pins": {
     "ai-models": {
       "repo": "https://github.com/artofrawr/claude-control",
@@ -136,6 +136,14 @@
       "_status": "active",
       "sha256": "6167ed595dde2a2aaef9736b49abcadfa4fc4e83fcb81bd28c45775697aee9f0"
     },
+    "humanizer": {
+      "repo": "https://github.com/blader/humanizer",
+      "tag": null,
+      "commit": "8b3a17889fbf12bedae20974a3c9f9de746ed754",
+      "pinned_at": "2026-05-20",
+      "_status": "active",
+      "sha256": "6e2bb628772b87750849d74e1176736c405f399ab4ebbf3f08e1ce26ed69852c"
+    },
     "i18n-localization": {
       "repo": "https://github.com/sickn33/antigravity-awesome-skills",
       "path": "skills/i18n-localization",
@@ -633,6 +641,9 @@
       {
         "path": "templates/skills-marketplace/huashu-design"
       },
+      {
+        "path": "templates/skills-marketplace/humanizer"
+      },
       {
         "path": "templates/skills-marketplace/i18n-localization"
       },

scripts/init-claude.sh

@@ -35,7 +35,7 @@ NC='\033[0m'
 # `raw.githubusercontent.com/.../v6.24.5/.../init-claude.sh`), leave
 # TK_TOOLKIT_REF unset and it inherits the bundled default below —
 # guaranteeing every file in the install comes from the same tag.
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/install-statusline.sh

@@ -33,7 +33,7 @@ done
 : "${YES}"  # silence shellcheck SC2034 — no-op stub today
 
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/install.sh

@@ -40,7 +40,7 @@ export TK_TOOLKIT_REF TK_USER_AGENT
 # Config
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`); TK_REPO_URL
 # remains the highest-priority override (full URL with ref baked in).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/migrate-to-complement.sh

@@ -57,7 +57,7 @@ CYAN='\033[0;36m'
 NC='\033[0m'
 
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/setup-council.sh

@@ -16,7 +16,7 @@ CYAN='\033[0;36m'
 NC='\033[0m'
 
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/setup-prompt-engineer.sh

@@ -18,7 +18,7 @@ NC='\033[0m'
 
 # TK_TOOLKIT_REF pins to a tag/SHA (default `main`). Mirrors setup-council.sh
 # to keep allowlist + curl conventions identical.
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 if ! [[ "$TK_TOOLKIT_REF" =~ ^[A-Za-z0-9._/-]+$ ]] || [[ "$TK_TOOLKIT_REF" == *..* ]]; then
     echo "Error: TK_TOOLKIT_REF must match [A-Za-z0-9._/-]+ and must not contain '..' (got: $TK_TOOLKIT_REF)" >&2
     exit 1

scripts/setup-security.sh

@@ -46,7 +46,7 @@ done
 : "${YES}"  # silence shellcheck SC2034 — YES consumed by future read blocks
 
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/uninstall.sh

@@ -104,7 +104,7 @@ fi
 
 # ───────── constants + log helpers ─────────
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

scripts/update-claude.sh

@@ -73,7 +73,7 @@ CYAN='\033[0;36m'
 NC='\033[0m'
 
 # Audit H5: TK_TOOLKIT_REF pins to a tag/SHA (default `main`).
-TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.49.0}"
+TK_TOOLKIT_REF="${TK_TOOLKIT_REF:-v6.50.0}"
 # Audit INF-MED-2 (2026-04-30 deep): allowlist guard — TK_TOOLKIT_REF flows
 # raw into curl URLs. Reject anything outside the tag/SHA charset, plus any
 # `..` traversal sequence. Tags / branches / SHAs do not contain `..`.

templates/skills-catalog.json

@@ -1,9 +1,9 @@
 {
   "schema_version": 1,
   "generated_from": "manifest.json:skills_pins",
-  "generated_at": "2026-05-18",
-  "toolkit_version": "6.47.0",
-  "skills_count": 22,
+  "generated_at": "2026-05-20",
+  "toolkit_version": "6.50.0",
+  "skills_count": 23,
   "skills": [
     {
       "name": "ai-models",
@@ -68,6 +68,13 @@
       "sha256": "6167ed595dde2a2aaef9736b49abcadfa4fc4e83fcb81bd28c45775697aee9f0",
       "install_cmd": "claude-skills install --from-github https://github.com/alchaincyf/huashu-design --commit 8e25b23709747a747b80bf4d3820def8f8054a4e"
     },
+    {
+      "name": "humanizer",
+      "upstream": "https://github.com/blader/humanizer",
+      "commit": "8b3a17889fbf12bedae20974a3c9f9de746ed754",
+      "sha256": "6e2bb628772b87750849d74e1176736c405f399ab4ebbf3f08e1ce26ed69852c",
+      "install_cmd": "claude-skills install --from-github https://github.com/blader/humanizer --commit 8b3a17889fbf12bedae20974a3c9f9de746ed754"
+    },
     {
       "name": "i18n-localization",
       "upstream": "https://github.com/sickn33/antigravity-awesome-skills",
@@ -180,4 +187,4 @@
     }
   ],
   "_note": "Auto-generated by scripts/generate-skills-catalog.sh. Do not edit by hand. Source of truth is manifest.json:skills_pins. This catalog is a compact declarative view for catalog-mode consumers; the toolkit's primary distribution remains the mirror under templates/skills-marketplace/. no-upstream-found pins (e.g., memo-skill) are intentionally omitted."
-}
+}