fix(claude-code-docs): vendor plugin into the marketplace

.agents/plugins/marketplace.json

@@ -559,8 +559,8 @@
     {
       "name": "claude-code-docs",
       "source": {
-        "source": "url",
-        "url": "https://github.com/pleaseai/claude-code-docs.git"
+        "source": "local",
+        "path": "./plugins/claude-code-docs"
       },
       "policy": {
         "installation": "AVAILABLE",

.claude-plugin/marketplace.json

@@ -772,10 +772,7 @@
           "filesRead": ["**/.claude-plugin/**", "**/.claude/settings.json", "**/hooks/hooks.json"]
         }
       },
-      "source": {
-        "source": "github",
-        "repo": "pleaseai/claude-code-docs"
-      },
+      "source": "./plugins/claude-code-docs",
       "homepage": "https://github.com/pleaseai/claude-code-docs"
     },
     {

.cursor-plugin/marketplace.json

@@ -199,6 +199,11 @@
       "source": "./plugins/fetch",
       "description": "Fetch web content in multiple formats - HTML, JSON, plain text, Markdown, readable articles, and YouTube transcripts"
     },
+    {
+      "name": "claude-code-docs",
+      "source": "./plugins/claude-code-docs",
+      "description": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs"
+    },
     {
       "name": "semble",
       "source": "./plugins/semble",

.release-please-manifest.json

@@ -54,6 +54,7 @@
   "plugins/graphite": "2.3.0",
   "plugins/bun": "1.4.0",
   "plugins/portone": "1.3.0",
+  "plugins/claude-code-docs": "1.0.0",
   "plugins/claude-md-management": "1.1.0",
   "plugins/cubic": "1.1.0",
   "plugins/fetch": "1.2.0",

README.md

@@ -80,11 +80,6 @@ Skills for the Cloudflare developer platform: Workers, Durable Objects, Agents S
 
 **Install:** `/plugin install cloudflare@pleaseai` | **Repository:** [cloudflare/skills](https://github.com/cloudflare/skills)
 
-#### Claude Code Docs
-Fetch Claude Code docs as markdown — rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs.
-
-**Install:** `/plugin install claude-code-docs@pleaseai` | **Repository:** [pleaseai/claude-code-docs](https://github.com/pleaseai/claude-code-docs)
-
 #### Run Verify
 Run and verify code changes by driving the real app — the `run`, `verify`, and `run-skill-generator` skills, generalized for Claude Code, Codex, and Antigravity.
 
@@ -498,6 +493,12 @@ Token-efficient browser automation CLI for coding agents — navigate, interact,
 
 **Install:** `/plugin install playwright-cli@pleaseai` | **Source:** [plugins/playwright-cli](https://github.com/pleaseai/claude-code-plugins/tree/main/plugins/playwright-cli)
 
+#### Claude Code Docs
+
+Fetch Claude Code docs as markdown — rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs. The docs mirror itself stays at [pleaseai/claude-code-docs](https://github.com/pleaseai/claude-code-docs).
+
+**Install:** `/plugin install claude-code-docs@pleaseai` | **Source:** [plugins/claude-code-docs](https://github.com/pleaseai/claude-code-plugins/tree/main/plugins/claude-code-docs)
+
 ## Quick Start
 
 The fastest way to get started — install the marketplace and let the plugin recommender auto-detect what you need:
@@ -600,7 +601,6 @@ Once the marketplace is added (or files copied), the following plugins are avail
 /plugin install vercel@pleaseai
 /plugin install gemini@pleaseai
 /plugin install cloudflare@pleaseai
-/plugin install claude-code-docs@pleaseai
 /plugin install firecrawl@pleaseai
 /plugin install ask@pleaseai
 
@@ -683,6 +683,7 @@ Once the marketplace is added (or files copied), the following plugins are avail
 /plugin install tosspayments@pleaseai
 /plugin install vercel-sandbox@pleaseai
 /plugin install playwright-cli@pleaseai
+/plugin install claude-code-docs@pleaseai
 ```
 
 ## What Are Claude Code Plugins?

plugins/claude-code-docs/.claude-plugin/plugin.json

@@ -0,0 +1,18 @@
+{
+  "name": "claude-code-docs",
+  "version": "1.0.0",
+  "description": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+  "author": {
+    "name": "PleaseAI",
+    "url": "https://github.com/pleaseai"
+  },
+  "homepage": "https://github.com/pleaseai/claude-code-docs",
+  "license": "MIT",
+  "keywords": [
+    "claude-code",
+    "docs",
+    "markdown",
+    "webfetch",
+    "hooks"
+  ]
+}

plugins/claude-code-docs/.codex-plugin/plugin.json

@@ -0,0 +1,32 @@
+{
+  "name": "claude-code-docs",
+  "version": "1.0.0",
+  "description": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+  "author": {
+    "name": "PleaseAI",
+    "url": "https://github.com/pleaseai"
+  },
+  "interface": {
+    "displayName": "Claude Code Docs",
+    "shortDescription": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+    "longDescription": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+    "developerName": "PleaseAI",
+    "category": "Tooling",
+    "capabilities": [
+      "Skill"
+    ],
+    "defaultPrompt": [
+      "Help me use Claude Code Docs for my current task."
+    ],
+    "websiteURL": "https://github.com/pleaseai/claude-code-docs"
+  },
+  "homepage": "https://github.com/pleaseai/claude-code-docs",
+  "license": "MIT",
+  "keywords": [
+    "claude-code",
+    "docs",
+    "markdown",
+    "webfetch",
+    "hooks"
+  ]
+}

plugins/claude-code-docs/.cursor-plugin/plugin.json

@@ -0,0 +1,23 @@
+{
+  "name": "claude-code-docs",
+  "displayName": "Claude Code Docs",
+  "description": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+  "version": "1.0.0",
+  "author": {
+    "name": "PleaseAI"
+  },
+  "category": "Tooling",
+  "homepage": "https://github.com/pleaseai/claude-code-docs",
+  "license": "MIT",
+  "keywords": [
+    "claude-code",
+    "docs",
+    "markdown",
+    "webfetch",
+    "hooks"
+  ],
+  "tags": [
+    "hooks",
+    "docs"
+  ]
+}

plugins/claude-code-docs/hooks.json

@@ -0,0 +1,17 @@
+{
+  "description": "Rewrites code.claude.com/docs page URLs to their .md source so WebFetch gets clean markdown instead of the HTML app shell",
+  "hooks": {
+    "PreToolUse": [
+      {
+        "matcher": "WebFetch",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "bash \"./hooks/rewrite-docs-url.sh\"",
+            "timeout": 5
+          }
+        ]
+      }
+    ]
+  }
+}

plugins/claude-code-docs/hooks/hooks.json

@@ -0,0 +1,17 @@
+{
+  "description": "Rewrites code.claude.com/docs page URLs to their .md source so WebFetch gets clean markdown instead of the HTML app shell",
+  "hooks": {
+    "PreToolUse": [
+      {
+        "matcher": "WebFetch",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "bash \"${CLAUDE_PLUGIN_ROOT}/hooks/rewrite-docs-url.sh\"",
+            "timeout": 5
+          }
+        ]
+      }
+    ]
+  }
+}

plugins/claude-code-docs/hooks/rewrite-docs-url.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+# PreToolUse hook for WebFetch: rewrite Claude Code docs page URLs to their
+# markdown source (https://code.claude.com/docs/en/agents -> .../agents.md).
+# Every docs page serves a clean markdown twin at <page>.md; fetching it skips
+# the HTML app shell entirely. Silent exit 0 = no opinion, normal flow.
+
+input=$(cat)
+url=$(jq -r '.tool_input.url // empty' <<<"$input")
+[ -z "$url" ] && exit 0
+
+case "$url" in
+  https://code.claude.com/docs/*) ;;
+  *) exit 0 ;;
+esac
+
+# Drop fragment and query: hooks#pretooluse-decision-control -> hooks
+path="${url%%#*}"
+path="${path%%\?*}"
+path="${path%/}"
+
+# Already a raw file (page .md, llms.txt, images, ...) - leave untouched.
+# Check the last path segment only; the domain always contains dots.
+case "${path##*/}" in
+  *.*) exit 0 ;;
+esac
+
+# Only rewrite real pages under a locale segment (/docs/<locale>/<page...>),
+# not the docs root itself
+rest="${path#https://code.claude.com/docs/}"
+case "$rest" in
+  */*) ;;
+  *) exit 0 ;;
+esac
+
+jq --arg url "${path}.md" '{
+  hookSpecificOutput: {
+    hookEventName: "PreToolUse",
+    permissionDecision: "allow",
+    permissionDecisionReason: ("Rewrote Claude Code docs URL to markdown source: " + $url),
+    updatedInput: (.tool_input | .url = $url)
+  }
+}' <<<"$input"

plugins/claude-code-docs/hooks/rewrite-docs-url.test.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+# Tests for rewrite-docs-url.sh. Run: bash hooks/rewrite-docs-url.test.sh
+set -u
+cd "$(dirname "$0")"
+pass=0 fail=0
+
+run() {
+  echo "$1" | bash ./rewrite-docs-url.sh
+}
+
+expect_rewrite() { # input-url expected-url
+  out=$(run "{\"tool_name\":\"WebFetch\",\"tool_input\":{\"url\":\"$1\",\"prompt\":\"p\"}}")
+  got=$(jq -r '.hookSpecificOutput.updatedInput.url // empty' <<<"$out")
+  if [ "$got" = "$2" ]; then pass=$((pass+1)); else fail=$((fail+1)); echo "FAIL: $1 -> '$got' (expected $2)"; fi
+}
+
+expect_silent() { # input-url
+  out=$(run "{\"tool_name\":\"WebFetch\",\"tool_input\":{\"url\":\"$1\",\"prompt\":\"p\"}}")
+  if [ -z "$out" ]; then pass=$((pass+1)); else fail=$((fail+1)); echo "FAIL: $1 should pass through, got: $out"; fi
+}
+
+expect_rewrite "https://code.claude.com/docs/en/agents" "https://code.claude.com/docs/en/agents.md"
+expect_rewrite "https://code.claude.com/docs/en/agents/" "https://code.claude.com/docs/en/agents.md"
+expect_rewrite "https://code.claude.com/docs/en/hooks#pretooluse-decision-control" "https://code.claude.com/docs/en/hooks.md"
+expect_rewrite "https://code.claude.com/docs/en/agent-sdk/python" "https://code.claude.com/docs/en/agent-sdk/python.md"
+expect_rewrite "https://code.claude.com/docs/en/hooks?foo=bar" "https://code.claude.com/docs/en/hooks.md"
+
+expect_silent "https://code.claude.com/docs/en/agents.md"
+expect_silent "https://code.claude.com/docs/llms.txt"
+expect_silent "https://code.claude.com/docs/llms-full.txt"
+expect_silent "https://code.claude.com/docs/en"
+expect_silent "https://code.claude.com/docs"
+expect_silent "https://example.com/docs/en/agents"
+expect_silent "https://docs.claude.com/en/api/overview"
+
+# prompt must survive updatedInput (it replaces the whole input object)
+out=$(run '{"tool_name":"WebFetch","tool_input":{"url":"https://code.claude.com/docs/en/agents","prompt":"summarize"}}')
+got=$(jq -r '.hookSpecificOutput.updatedInput.prompt' <<<"$out")
+if [ "$got" = "summarize" ]; then pass=$((pass+1)); else fail=$((fail+1)); echo "FAIL: prompt lost: $got"; fi
+
+echo "passed: $pass, failed: $fail"
+[ "$fail" -eq 0 ]

plugins/claude-code-docs/plugin.json

@@ -0,0 +1,19 @@
+{
+  "$schema": "https://antigravity.google/schemas/v1/plugin.json",
+  "name": "claude-code-docs",
+  "version": "1.0.0",
+  "description": "Fetch Claude Code docs as markdown - rewrites code.claude.com/docs page URLs to their .md source before WebFetch runs",
+  "author": {
+    "name": "PleaseAI",
+    "url": "https://github.com/pleaseai"
+  },
+  "homepage": "https://github.com/pleaseai/claude-code-docs",
+  "license": "MIT",
+  "keywords": [
+    "claude-code",
+    "docs",
+    "markdown",
+    "webfetch",
+    "hooks"
+  ]
+}

plugins/claude-code-docs/skills/claude-code-docs/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: claude-code-docs
+description: Answer questions about Claude Code (features, hooks, plugins, skills, MCP, Agent SDK, settings, CLI) from the official docs. Uses the ask CLI to resolve a version-pinned local mirror and csp for semantic search, falling back to grep and WebFetch. Use when the user asks how Claude Code works, what a configuration field does, or whether Claude Code supports something.
+argument-hint: [question]
+arguments: [question]
+---
+
+Answer this question about Claude Code using the official documentation:
+
+**$question**
+
+## How to find the answer
+
+Work through these steps, falling back gracefully when a tool is missing.
+
+### 1. Resolve the docs mirror
+
+```bash
+DOCS="$(ask src github:pleaseai/claude-code-docs)/docs"
+```
+
+`ask src` prints the root of a cached checkout of the docs mirror (clones on first use). If the `ask` CLI is not on PATH, run it via npx/bunx instead:
+
+```bash
+DOCS="$(npx -y @pleaseai/ask src github:pleaseai/claude-code-docs)/docs"
+```
+
+Only fall back to step 4 (WebFetch) when neither works, such as offline sandboxes without npm registry access.
+
+### 2. Search for relevant pages
+
+Prefer semantic search with csp, scoped to the per-page markdown:
+
+```bash
+csp search "$question" "$DOCS" --content docs --top-k 8
+```
+
+Or equivalently via ask (requires csp on PATH):
+
+```bash
+ask search github:pleaseai/claude-code-docs "$question" --content docs --top-k 8
+```
+
+If csp is not on PATH, run it via npx/bunx:
+
+```bash
+npx -y @pleaseai/csp search "$question" "$DOCS" --content docs --top-k 8
+```
+
+If that fails too, fall back to grep over the mirror — search for 2-3 distinct keyword variants, not one long phrase:
+
+```bash
+grep -ril "<keyword>" "$DOCS"
+```
+
+The manifest at `$DOCS/../llms.txt` lists every page with a one-line summary; reading it is a fast way to shortlist pages when keyword search is noisy.
+
+### 3. Read and answer
+
+Read the top matching pages (whole pages — they are small). Ground every claim in what the pages actually say:
+
+- Answer in the user's language; keep code identifiers, field names, and flags verbatim.
+- Cite each source page with its canonical URL: a mirror file `$DOCS/hooks.md` corresponds to `https://code.claude.com/docs/en/hooks`.
+- Quote exact field names, JSON shapes, and version requirements (`min-version` markers) when they matter.
+- If the docs do not answer the question, say so explicitly — do not fill gaps from memory.
+
+### 4. Fallback: WebFetch
+
+If the `ask` CLI is unavailable, fetch `https://code.claude.com/docs/llms.txt`, pick the relevant page URLs from the manifest, and fetch each as `<page-url>.md` (this plugin's hook rewrites page URLs to `.md` automatically). Then answer as in step 3.