@contentrain/skills 0.2.1 → 0.4.0

package/README.md

@@ -1,76 +1,115 @@
 # `@contentrain/skills`
 
 [![npm version](https://img.shields.io/npm/v/%40contentrain%2Fskills?label=%40contentrain%2Fskills)](https://www.npmjs.com/package/@contentrain/skills)
+[![Agent Skills](https://img.shields.io/badge/Agent_Skills-15_skills-8B5CF6)](https://agentskills.io)
 [![GitHub source](https://img.shields.io/badge/source-Contentrain%2Fai-181717?logo=github)](https://github.com/Contentrain/ai/tree/main/packages/skills)
+[![Docs](https://img.shields.io/badge/docs-ai.contentrain.io-0f172a)](https://ai.contentrain.io/packages/skills)
 
 Workflow skills and framework guides for Contentrain-aware AI agents.
 
-This package follows the [Agent Skills standard](https://agentskills.io) for progressive disclosure: each skill has a `SKILL.md` (loaded on activation) and optional `references/` (loaded on demand).
+Start here:
+
+- [2-minute product demo](https://ai.contentrain.io/demo)
+- [Rules & skills docs](https://ai.contentrain.io/packages/rules)
+- [Framework integration guide](https://ai.contentrain.io/guides/frameworks)
+
+This package follows the [Agent Skills standard](https://agentskills.io) for **progressive disclosure**: each skill has a `SKILL.md` (loaded on activation) and optional `references/*.md` (loaded on demand).
 
 ## Install
 
+### Via npm (programmatic access to catalogs and constants)
+
 ```bash
 pnpm add @contentrain/skills
 ```
 
-Or install directly via the skills CLI:
+### Via skills CLI (install skills into your AI agent)
 
 ```bash
-npx skills add contentrain/contentrain-ai --skill='*'
+# Install all 15 skills
+npx skills add Contentrain/ai/packages/skills
+
+# Install a specific skill
+npx skills add Contentrain/ai/packages/skills --skill contentrain-normalize
+
+# Install to a specific agent
+npx skills add Contentrain/ai/packages/skills --agent claude-code
+
+# List available skills
+npx skills add Contentrain/ai/packages/skills --list
 ```
 
+Works with Claude Code, Cursor, Windsurf, GitHub Copilot, OpenAI Codex, Gemini CLI, and 40+ other agents.
+
 ## What It Contains
 
 ### Agent Skills (standard format)
 
-Published under `skills/`:
+Published under `skills/` — 15 production skills:
 
 | Skill | Description |
 |-------|-------------|
-| `contentrain` | Core architecture, MCP tools, content formats |
-| `contentrain-normalize` | Two-phase normalize (extract + reuse) |
-| `contentrain-quality` | Content quality, SEO, accessibility, media |
-| `contentrain-sdk` | @contentrain/query SDK usage (local + CDN) |
-| `contentrain-content` | Content CRUD operations |
-| `contentrain-model` | Model creation/update |
-| `contentrain-init` | Project initialization |
-| `contentrain-bulk` | Batch operations |
-| `contentrain-validate-fix` | Validation and auto-fix |
-| `contentrain-review` | Content quality review |
-| `contentrain-translate` | Multi-locale translation |
-| `contentrain-generate` | SDK client generation |
-| `contentrain-serve` | Local review/normalize UI |
-| `contentrain-diff` | Branch content diffs |
-| `contentrain-doctor` | Project health diagnostics |
+| `contentrain` | Core architecture, MCP tools, content formats, i18n, security |
+| `contentrain-normalize` | Two-phase normalize (extract hardcoded strings + patch source files) |
+| `contentrain-quality` | Content quality, SEO, accessibility, media optimization |
+| `contentrain-sdk` | Query SDK usage (#contentrain imports, QueryBuilder, type-safe access) |
+| `contentrain-content` | Create and manage content entries for existing models |
+| `contentrain-model` | Design and save model definitions |
+| `contentrain-init` | Initialize a new Contentrain project |
+| `contentrain-bulk` | Batch operations on content entries |
+| `contentrain-validate-fix` | Validate content and auto-fix structural issues |
+| `contentrain-review` | Review content changes before publishing |
+| `contentrain-translate` | Multi-locale translation workflows |
+| `contentrain-generate` | Generate the typed SDK client from models |
+| `contentrain-serve` | Start the local review and normalize UI |
+| `contentrain-diff` | View content diffs between branches |
+| `contentrain-doctor` | Diagnose project health issues |
 
 Each skill directory contains:
+
 ```
 skills/{name}/
-├── SKILL.md           # Instructions (< 500 lines)
+├── SKILL.md           # Instructions (< 500 lines, quick reference)
 └── references/        # Detailed docs (loaded on demand)
-    └── *.md
+    └── *.md           # Deep dives: architecture, patterns, examples
 ```
 
+Example references: `contentrain/references/mcp-tools.md`, `contentrain-normalize/references/extraction.md`, `contentrain-quality/references/seo.md`.
+
 ### Workflow skills (backward compat)
 
-Published under `workflows/*` — flat markdown files from previous versions. Still functional.
+Published under `workflows/` — flat markdown files retained for compatibility:
+
+- `contentrain-init.md`, `contentrain-model.md`, `contentrain-content.md`, `contentrain-bulk.md`
+- `contentrain-normalize.md`, `contentrain-validate-fix.md`, `contentrain-review.md`
+- `contentrain-diff.md`, `contentrain-doctor.md`, `contentrain-serve.md`
+- `contentrain-generate.md`, `contentrain-translate.md`
+
+Still fully functional; new projects should prefer the `skills/` structure.
 
 ### Framework guides
 
-Published under `frameworks/*`:
+Published under `frameworks/` — per-framework integration patterns:
 
-- `nuxt.md`, `next.md`, `astro.md`, `sveltekit.md`, `vue.md`, `react.md`, `expo.md`, `react-native.md`, `node.md`
+- `nuxt.md`, `next.md`, `astro.md`, `sveltekit.md` (meta frameworks)
+- `react.md`, `vue.md` (UI libraries)
+- `expo.md`, `react-native.md` (mobile)
+- `node.md` (backend)
 
 ## Public Exports
 
 ```ts
 import { AGENT_SKILLS, WORKFLOW_SKILLS, FRAMEWORK_GUIDES } from '@contentrain/skills'
 
-// Agent Skills catalog (name + description for Tier 1 discovery)
-console.log(AGENT_SKILLS)
+// Discover skills: name + description for agent activation
+AGENT_SKILLS.forEach(({ name, description }) => {
+  console.log(`${name}: ${description}`)
+})
+
+// Backward compat: flat workflow list
+console.log(WORKFLOW_SKILLS.length) // 12
 
-// Backward compat
-console.log(WORKFLOW_SKILLS)
+// Framework discovery
 console.log(FRAMEWORK_GUIDES)
 ```
 
@@ -78,15 +117,24 @@ console.log(FRAMEWORK_GUIDES)
 
 | Tier | What's Loaded | When | Token Cost |
 |------|--------------|------|------------|
-| 1. Catalog | `name` + `description` | Session start | ~50 tokens/skill |
+| 1. Catalog | `name` + `description` (AGENT_SKILLS) | Session start | ~50 tokens/skill |
 | 2. Instructions | Full `SKILL.md` body | Skill activated | < 5000 tokens |
-| 3. References | `references/*.md` | Agent needs detail | Varies |
+| 3. References | `references/*.md` files | Agent needs detail | Varies (50-500 tokens/file) |
+
+This reduces always-loaded context from thousands of lines to just the essentials plus catalog.
+
+## Parity with `@contentrain/mcp`
+
+`@contentrain/skills` is kept in lockstep with the MCP tool registry via cross-package parity tests (`tests/mcp-parity.test.ts`):
+
+- `skills/contentrain/references/mcp-tools.md` must have an `### <tool>` heading for every tool in the MCP `TOOL_NAMES` registry (currently 17).
+- Key skills (normalize, translate) must not reference legacy `contentrain/{operation}/...` branch prefixes — MCP now emits `cr/*`.
 
-This reduces always-loaded context from ~4,700 lines to ~86 lines (essential guardrails only).
+When MCP's surface changes, these tests fail until the skill docs catch up.
 
 ## IDE Support
 
-Skills are installed by `contentrain init` for detected IDEs:
+Skills are installed by `contentrain init` (and re-applied with `contentrain skills --update`) for detected IDEs:
 
 | IDE | Rules Dir | Skills Dir |
 |-----|-----------|------------|

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@contentrain/skills",
-  "version": "0.2.1",
+  "version": "0.4.0",
   "license": "MIT",
   "description": "AI agent skills for Contentrain — workflow procedures, framework integration guides",
   "type": "module",
@@ -9,7 +9,7 @@
     "url": "git+https://github.com/Contentrain/ai.git",
     "directory": "packages/skills"
   },
-  "homepage": "https://github.com/Contentrain/ai/tree/main/packages/skills",
+  "homepage": "https://ai.contentrain.io/packages/rules#skills",
   "bugs": {
     "url": "https://github.com/Contentrain/ai/issues"
   },
@@ -17,8 +17,14 @@
     "contentrain",
     "ai",
     "skills",
+    "agent-skills",
     "mcp",
     "workflow",
+    "claude-code",
+    "cursor",
+    "windsurf",
+    "copilot",
+    "codex",
     "vue",
     "react",
     "nuxt",
@@ -54,7 +60,8 @@
     "@types/node": "^22.0.0",
     "tsdown": "^0.21.0",
     "typescript": "^5.7.0",
-    "vitest": "^3.0.0"
+    "vitest": "^3.0.0",
+    "@contentrain/mcp": "1.3.0"
   },
   "scripts": {
     "build": "tsdown src/index.ts --format esm,cjs --dts",

package/skills/contentrain/SKILL.md

@@ -10,6 +10,34 @@ metadata:
 
 Contentrain is a Git-based, AI-first content management system. Content lives in `.contentrain/` as JSON and Markdown files, managed through MCP tools. This skill covers architecture, content formats, and tool usage.
 
+## Ecosystem Overview
+
+Contentrain consists of 6 packages that work together:
+
+| Package | Role | How agent uses it |
+|---|---|---|
+| @contentrain/mcp | 17 MCP tools (scan, apply, validate, merge, doctor...) | MCP tool calls |
+| contentrain (CLI) | init, serve, generate, doctor, diff, status | Shell commands |
+| @contentrain/types | Shared TypeScript contracts | Type safety |
+| @contentrain/query | Generated SDK client (Prisma-pattern) | `import from '#contentrain'` |
+| @contentrain/rules | Behavioral guardrails (~86 lines, always-loaded) | Auto-loaded by IDE |
+| @contentrain/skills | Procedural workflows (this file) | On-demand by agent |
+
+### CLI Serve — Review & Approval Bridge
+
+For workflows requiring human review (normalize, content review):
+
+```bash
+contentrain serve  # http://localhost:3333
+```
+
+| Page | When to use |
+|---|---|
+| /normalize | Extraction/reuse preview and approval |
+| /branches | Merge or delete pending branches |
+| /validate | Review validation errors visually |
+| /content | Browse models and entries |
+
 ## Architecture: Agent vs MCP
 
 The system separates **intelligence** (agent) from **infrastructure** (MCP tools).
@@ -94,6 +122,7 @@ slug: getting-started
 | `contentrain_apply` | Apply normalize operation (extract or reuse) |
 | `contentrain_validate` | Validate content against model schemas |
 | `contentrain_submit` | Push contentrain/* branches to remote |
+| `contentrain_merge` | Merge a review-mode branch into contentrain locally |
 | `contentrain_bulk` | Run batch operations on existing content |
 
 ## Calling Sequences
@@ -121,6 +150,18 @@ apply(reuse, dry_run:true) --> [review] --> apply(reuse, dry_run:false)
   --> validate --> submit --> [repeat per model]
 ```
 
+## MUST Rules
+
+These are enforced by MCP — violating them causes errors. Know them upfront:
+
+- MUST call `contentrain_status` before any write operation on existing projects
+- MUST call `contentrain_describe_format` before creating models or content
+- MUST use `dry_run: true` before `contentrain_apply` execute
+- MUST call `contentrain_validate` before `contentrain_submit`
+- MUST run `npx contentrain generate` after model/content changes that affect SDK
+- MUST NOT write to `.contentrain/meta/`, `.contentrain/client/`, `.contentrain/context.json` directly
+- MUST NOT auto-merge normalize branches (always review mode)
+
 ## Key Guardrails
 
 1. **Always call `contentrain_status` first** when working with an existing project
@@ -147,3 +188,10 @@ apply(reuse, dry_run:true) --> [review] --> apply(reuse, dry_run:false)
 | [i18n](references/i18n.md) | Internationalization quality rules |
 | [Security](references/security.md) | XSS prevention, secret detection, URL validation |
 | [Workflow](references/workflow.md) | Git workflow, branch naming, conflict resolution |
+
+## Related Skills
+
+- **contentrain-normalize** — Two-phase extraction and source patching
+- **contentrain-serve** — Browser-based review and approval UI
+- **contentrain-generate** — SDK client generation
+- **contentrain-sdk** — @contentrain/query usage

package/skills/contentrain/references/mcp-pipelines.md

@@ -147,7 +147,7 @@ contentrain_submit                              # Push (always review mode)
 
 - A dedicated `contentrain` branch is the single source of truth for content state, created at init and protected from deletion
 - Every write operation creates a temporary worktree on a new feature branch forked from `contentrain`
-- Branch naming: `contentrain/{operation}/{model}/{timestamp}` (locale included when applicable)
+- Branch naming: `cr/{operation}/{model}/{timestamp}` (locale included when applicable)
 - Do not create branches manually. MCP handles Git transactions
 - Developer's working tree is never mutated during MCP operations (no stash, no checkout, no merge on the developer's tree)
 - context.json is committed together with content changes, not as a separate commit

package/skills/contentrain/references/mcp-tools.md

@@ -269,13 +269,35 @@ Validate project content against model schemas.
 
 ### contentrain_submit
 
-Push contentrain/* branches to remote.
+Push `cr/*` feature branches to remote.
 
 | Parameter | Type | Required | Description |
 |-----------|------|----------|-------------|
-| `branches` | string[] | No | Specific branches to push (omit for all contentrain/* branches) |
+| `branches` | string[] | No | Specific branches to push (omit for all `cr/*` branches) |
 | `message` | string | No | Optional message for the push operation |
 
+### contentrain_merge
+
+Merge a single review-mode `cr/*` branch into the content-tracking `contentrain` branch and advance the base branch via `update-ref`. Runs the worktree transaction with selective sync — dirty files in the developer's working tree are preserved rather than overwritten.
+
+| Parameter | Type | Required | Description |
+|-----------|------|----------|-------------|
+| `branch` | string | Yes | Feature branch name (must start with `cr/`) |
+
+Returns `{ action, commit, sync }` — `sync.skipped[]` lists files the selective sync skipped because the developer has uncommitted changes. The CLI surfaces this as a warning.
+
+## Doctor Tools
+
+### contentrain_doctor
+
+Structured project health report: git install, Node version, `.contentrain/` structure, model parse, orphan content, pending branch pressure, SDK client freshness. Read-only. Local-filesystem only (`localWorktree` capability — unavailable over remote providers).
+
+| Parameter | Type | Required | Description |
+|-----------|------|----------|-------------|
+| `usage` | boolean | No | Run the heavier usage-analysis branch (unused content keys, duplicate dictionary values, locale coverage). Default `false` |
+
+Returns `{ checks[], summary: { total, passed, failed, warnings }, usage? }`. Each check carries `name`, `pass`, `detail`, optional `severity: 'error' | 'warning' | 'info'`.
+
 ## Bulk Tools
 
 ### contentrain_bulk

package/skills/contentrain/references/workflow.md

@@ -45,21 +45,21 @@ Contentrain supports two workflow modes, configured in `.contentrain/config.json
 
 ## 2. Branch Naming Convention
 
-All Contentrain branches follow a strict naming pattern:
+All Contentrain feature branches follow a strict naming pattern:
 
 ```
-contentrain/{operation}/{model}/{locale}/{timestamp}
+cr/{operation}/{model}/{locale}/{timestamp}-{suffix}
 ```
 
 ### Examples
 
 | Scenario | Branch Name |
 |----------|-------------|
-| Content update | `contentrain/content/blog-post/en/1710300000` |
-| Model creation | `contentrain/model/team-member/1710300000` |
-| Normalize extraction | `contentrain/normalize/extract/blog/1710300000` |
-| Normalize reuse | `contentrain/normalize/reuse/marketing-hero/en/1710300000` |
-| Scaffold | `contentrain/new/scaffold-landing/en/1710300000` |
+| Content update | `cr/content/blog-post/en/1710300000-a1b2` |
+| Model creation | `cr/model/team-member/1710300000-c3d4` |
+| Normalize extraction | `cr/normalize/extract/blog/1710300000-e5f6` |
+| Normalize reuse | `cr/normalize/reuse/marketing-hero/en/1710300000-0789` |
+| Scaffold | `cr/new/scaffold-landing/en/1710300000-1234` |
 
 ### Rules
 

package/skills/contentrain-bulk/SKILL.md

@@ -97,3 +97,10 @@ Report:
 - locale or entry count affected
 - validation outcome if run
 - submit status
+
+## Related Skills
+
+- **contentrain-content** — Single-entry content operations
+- **contentrain-validate-fix** — Validate after bulk changes
+- **contentrain-generate** — Regenerate SDK after bulk locale copy
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-content/SKILL.md

@@ -18,6 +18,16 @@ The user wants to create new content, add entries to a collection, populate a si
 
 ---
 
+## MUST Rules
+
+- MUST call `contentrain_describe` before writing content (know the schema)
+- MUST provide ALL required fields in entries
+- MUST create entries for ALL supported locales when model has `i18n: true`
+- MUST NOT set system fields (id, slug, createdAt, updatedAt, status, source)
+- MUST run `npx contentrain generate` after adding new models or changing fields
+
+---
+
 ## Steps
 
 ### 1. Check Project State
@@ -53,6 +63,7 @@ If the user's request does not match an existing model, offer to create one firs
 
 - Check `.contentrain/vocabulary.json` for canonical terms. All content must use these terms consistently across locales. Do not use alternative spellings or synonyms for vocabulary-defined terms.
 - Check `.contentrain/context.json` for tone conventions (professional, casual, technical).
+- **For dictionaries:** Before creating any new key, call `contentrain_content_list` on the target model and scan existing values. If the value you intend to write already exists under a different key, REUSE the existing key instead of creating a new one. MCP will also warn you via `advisories` in the save response if duplicates are detected.
 
 ### 5. Generate Content
 
@@ -114,7 +125,17 @@ contentrain_content_save({
 - Prefer batch mode -- send multiple entries in a single call when possible.
 - Locale defaults to the project's default locale if omitted, but explicit locale is recommended.
 
-### 9. Handle i18n Completeness
+### 9. Review Save Advisories
+
+After `contentrain_content_save`, check the response for an `advisories` field. If present:
+
+- Review each advisory for duplicate value warnings.
+- If a duplicate is flagged, consider whether to:
+  - Remove the new key and reuse the existing one instead.
+  - Keep both keys if they serve semantically different purposes (document the reason).
+- Report any advisories to the user for their decision.
+
+### 10. Handle i18n Completeness
 
 If the project has multiple supported locales and the model has `i18n: true`:
 
@@ -128,7 +149,7 @@ If the project has multiple supported locales and the model has `i18n: true`:
 
 If the user defers translations, note incomplete locales in the final summary.
 
-### 10. Validate
+### 11. Validate
 
 Call `contentrain_validate` to check all changes:
 
@@ -139,7 +160,9 @@ Call `contentrain_validate` to check all changes:
 
 Fix any errors reported. Acknowledge any warnings.
 
-### 11. Submit
+For large content operations (10+ entries), also recommend running `contentrain doctor --usage` to detect unused keys and duplicate values across the project.
+
+### 12. Submit
 
 Call `contentrain_submit` to push branches to remote:
 
@@ -150,7 +173,17 @@ Call `contentrain_submit` to push branches to remote:
 
 Report the Studio URL if provided in the tool response -- the user can review and approve content there.
 
-### 12. Final Summary
+### After Save: SDK Regeneration
+
+When new models are created or fields change, regenerate the SDK client:
+
+```bash
+npx contentrain generate
+```
+
+This updates TypeScript types for the new content. See: **contentrain-generate** skill.
+
+### 13. Final Summary
 
 Report to the user:
 
@@ -160,3 +193,10 @@ Report to the user:
 - The submission status (auto-merged or pending review).
 - Studio link for visual review (if available).
 - Next steps: run `/contentrain-generate` to update SDK types if models changed.
+
+## Related Skills
+
+- **contentrain-serve** — Browser-based review for content branches
+- **contentrain-generate** — SDK client regeneration after content changes
+- **contentrain-quality** — Content quality, SEO, and accessibility rules
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-diff/SKILL.md

@@ -60,3 +60,18 @@ Summarize:
 - action taken
 - base branch affected
 - any follow-up branch still pending
+
+## Browser Alternative
+
+For visual branch review, use CLI serve:
+
+```bash
+contentrain serve
+```
+
+Navigate to http://localhost:3333/branches to see all pending branches with diffs, and merge or delete them visually.
+
+## Related Skills
+
+- **contentrain-serve** — Browser-based branch review and merge
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-doctor/SKILL.md

@@ -60,3 +60,10 @@ Recommended order:
 ### 5. Re-run Doctor
 
 After changes, run `contentrain doctor` again to confirm the project is healthy.
+
+## Related Skills
+
+- **contentrain-init** — Re-initialize if project structure is broken
+- **contentrain-validate-fix** — Fix content/model validation errors
+- **contentrain-generate** — Regenerate SDK if client is stale
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-generate/SKILL.md

@@ -181,3 +181,10 @@ Report to the user:
 - TypeScript types: available via `index.d.ts`.
 - Next steps: import `query` from `#contentrain` and start querying content.
 - Reminder: re-run `contentrain generate` after model or content changes.
+
+## Related Skills
+
+- **contentrain** — Core architecture and MCP tool catalog
+- **contentrain-normalize** — Normalize workflow requires SDK regen after Phase 2
+- **contentrain-content** — Content changes require SDK regen for type updates
+- **contentrain-sdk** — @contentrain/query usage after generation

package/skills/contentrain-generate/references/generated-client.md

@@ -110,11 +110,13 @@ import { query, singleton, dictionary, document } from '#contentrain'
 ```ts
 const posts = query('blog-post')
   .locale('en')               // set locale
-  .where('status', 'published') // exact match filter
+  .where('status', 'published') // equality filter (shorthand)
+  .where('views', 'gt', 100)  // operator filter: eq|ne|gt|gte|lt|lte|in|contains
   .sort('date', 'desc')       // sort by field
   .limit(10)                  // limit results
   .offset(5)                  // skip results
   .include('author', 'tags')  // resolve relation fields (1 level deep)
+  .count()                    // --> number (count of matches)
   .all()                      // --> T[] (returns array)
   .first()                    // --> T | undefined (first match)
 ```
@@ -143,22 +145,22 @@ Parameterized templates use `{placeholder}` syntax in dictionary values. The `ge
 ```ts
 const article = document('blog-article')
   .locale('en')
-  .where('category', 'tech')
+  .where('category', 'tech')           // equality shorthand
+  .where('title', 'contains', 'Guide') // operator syntax
   .include('author')          // resolve relations in frontmatter
   .bySlug('getting-started')  // --> T | undefined (find by slug)
 
-const docs = document('doc-page').locale('en').all()    // --> T[]
-const first = document('doc-page').locale('en').first()  // --> T | undefined
+const docs = document('doc-page').locale('en').all()     // --> T[]
+const first = document('doc-page').locale('en').first()   // --> T | undefined
+const total = document('doc-page').locale('en').count()   // --> number
 ```
 
 ### DOES NOT EXIST -- never use these
 
-- `.filter()` -- use `.where(field, value)` instead
+- `.filter()` -- use `.where(field, value)` or `.where(field, op, value)` instead
 - `.byId()` -- use `.where('id', value).first()` instead
-- `.count()` -- use `.all().length` instead
 - `dictionary().all()` -- use `.get()` instead
 - Queries are SYNC -- do not use `await` with `query()`, `singleton()`, `dictionary()`, or `document()`
-- `.where('field', 'eq', value)` -- just `.where('field', value)`
 - `.get()` on QueryBuilder -- use `.all()` or `.first()`
 
 ---

package/skills/contentrain-init/SKILL.md

@@ -96,4 +96,21 @@ Report to the user:
 - What was created in `.contentrain/`.
 - Which models were set up.
 - How to import content in their code (show the `#contentrain` import).
-- Next steps: create content, run `npx contentrain generate`, start the dev server.
+- **Next step: Run `npx contentrain generate` to create the typed SDK client.** Without this, `#contentrain` imports won't work. See: **contentrain-generate** skill.
+- Start the dev server and verify imports work.
+
+### Post-Init: Start Review UI
+
+After initialization, start the local review UI to explore the project:
+
+```bash
+contentrain serve
+```
+
+This opens http://localhost:3333 where you can browse models, content, and validation results.
+
+## Related Skills
+
+- **contentrain-serve** — Local review UI
+- **contentrain-generate** — SDK client generation
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-model/SKILL.md

@@ -139,3 +139,10 @@ Report:
 - field count
 - any relation targets
 - whether content migration is needed next
+
+## Related Skills
+
+- **contentrain-content** — Create content for this model after defining it
+- **contentrain-validate-fix** — Validate model and content after changes
+- **contentrain-generate** — Regenerate SDK after model changes
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-normalize/SKILL.md

@@ -17,6 +17,42 @@ Phase 1 alone is valuable: content becomes manageable in Studio, translatable, a
 
 ---
 
+## MUST Rules
+
+- MUST scan before extract (`contentrain_scan` → `contentrain_apply`)
+- MUST `dry_run: true` before `dry_run: false` for every `contentrain_apply` call
+- MUST merge Phase 1 branch before starting Phase 2
+- MUST run `npx contentrain generate` after Phase 2 completes
+- MUST NOT reuse without scoped model or domain (whole-project patching is blocked)
+- MUST NOT patch `.contentrain/` files via reuse (content files are read-only for reuse)
+- MUST NOT exceed 100 patches per `contentrain_apply` call
+
+## Transport Requirements
+
+Normalize (`contentrain_scan` and `contentrain_apply`) requires **local
+disk access** — AST scanners walk the source tree and patch files in
+place. It runs only on a `LocalProvider` (stdio transport, or an HTTP
+transport configured with a `LocalProvider`).
+
+Remote providers (`GitHubProvider`, `GitLabProvider`, future
+`BitbucketProvider`) expose `astScan: false`, `sourceRead: false`, and
+`sourceWrite: false`. Calling these tools over a remote provider
+returns a uniform capability error:
+
+```json
+{
+  "error": "contentrain_scan requires local filesystem access.",
+  "capability_required": "astScan",
+  "hint": "This tool is unavailable when MCP is driven by a remote provider. Use a LocalProvider or the stdio transport."
+}
+```
+
+If the agent is driving a remote-only MCP session, normalize must run
+in a separate local-checkout session before the extracted content
+branch is pushed.
+
+---
+
 ## Two-Phase Architecture
 
 | Aspect | Phase 1: Extraction | Phase 2: Reuse |
@@ -24,7 +60,7 @@ Phase 1 alone is valuable: content becomes manageable in Studio, translatable, a
 | Purpose | Pull content from source to `.contentrain/` | Patch source files with content references |
 | Scope | Full project scan | Per model or per domain |
 | Source files modified | No | Yes |
-| Branch pattern | `contentrain/normalize/extract/{domain}/{timestamp}` | `contentrain/normalize/reuse/{model}/{locale}/{timestamp}` |
+| Branch pattern | `cr/normalize/extract/{domain}/{timestamp}` | `cr/normalize/reuse/{model}/{locale}/{timestamp}` |
 | Prerequisite | Initialized `.contentrain/` | Completed extraction (content exists in `.contentrain/`) |
 | Workflow mode | Always `review` | Always `review` |
 | Standalone value | Yes -- content is manageable in Studio immediately | Depends on Phase 1 |
@@ -97,7 +133,7 @@ Call `contentrain_apply(mode: "extract", dry_run: true)` to generate a preview.
 
 ### 6. Execute Extraction
 
-After user approval, call `contentrain_apply(mode: "extract", dry_run: false)`. This creates model definitions and content files in `.contentrain/` on a `contentrain/normalize/extract/{timestamp}` branch. Source files are NOT modified.
+After user approval, call `contentrain_apply(mode: "extract", dry_run: false)`. This creates model definitions and content files in `.contentrain/` on a `cr/normalize/extract/{timestamp}` branch. Source files are NOT modified.
 
 ### 7. Validate and Submit
 
@@ -105,6 +141,52 @@ Call `contentrain_validate` to check schema compliance, i18n completeness, and d
 
 Tell the user: "Phase 1 complete. Content is now in Contentrain and can be managed, translated, and published from Studio. When ready, proceed with Phase 2 to update source files."
 
+### Step 5: User Review
+
+Two paths — agent chooses based on context:
+
+**Path A — Browser Review (recommended for visual diff):**
+1. Ensure `contentrain serve` is running (or ask user to start it)
+2. Summarize the dry_run output to the user
+3. Direct user to http://localhost:3333/normalize
+4. Explain: "On this page you can review the extraction plan and Approve or Reject it."
+5. Wait for user decision
+
+**Path B — Terminal Review:**
+1. Show dry_run output to user
+2. Ask for explicit approval
+3. If approved: `contentrain_apply(mode: 'extract', dry_run: false)` → execute
+4. `contentrain_validate` → `contentrain_submit`
+5. `contentrain_merge(branch: 'cr/normalize/extract/...', confirm: true)` to merge locally
+
+### Step 6: Merge Phase 1 Branch
+
+Phase 2 CANNOT start until Phase 1 branch is merged. Three options:
+- **Browser:** localhost:3333/branches → click Merge
+- **MCP Tool:** `contentrain_merge(branch: 'cr/normalize/extract/...', confirm: true)`
+- **Git platform:** Create PR → review → merge
+
+Verify with `contentrain_status` — check that extract branch is no longer in unmerged list.
+
+### Step 7: Phase 2 Prerequisite Check
+
+Before starting reuse:
+1. Call `contentrain_status` and verify extract branch is merged
+2. If not merged → inform user and wait
+3. If merged → proceed to Phase 2 reuse
+
+### Step 8: SDK Regeneration (after Phase 2)
+
+After all reuse operations are complete, source files now use `#contentrain` imports.
+The SDK client MUST be regenerated:
+
+```bash
+npx contentrain generate
+```
+
+Without this, `#contentrain` imports will fail at build time.
+See: **contentrain-generate** skill for details.
+
 ---
 
 ## Phase 2: Reuse (Step-by-Step)
@@ -125,7 +207,7 @@ Call `contentrain_apply(mode: "reuse", scope: { model: "<model-id>" }, patches:
 
 ### 4. Execute Reuse
 
-After user confirmation, call `contentrain_apply(mode: "reuse", scope: { model: "<model-id>" }, patches: [...], dry_run: false)`. This patches source files and creates a `contentrain/normalize/reuse/{model}/{timestamp}` branch.
+After user confirmation, call `contentrain_apply(mode: "reuse", scope: { model: "<model-id>" }, patches: [...], dry_run: false)`. This patches source files and creates a `cr/normalize/reuse/{model}/{timestamp}` branch.
 
 ### 5. Validate and Submit
 
@@ -183,3 +265,10 @@ Ask the user which model or domain to process next. Repeat steps 1-5 for each re
 - [Extraction Details](references/extraction.md) -- Phase 1 extraction rules, parameters, and examples
 - [Reuse Details](references/reuse.md) -- Phase 2 replacement expressions by stack and patching rules
 - [What Is Content](references/what-is-content.md) -- Heuristics for identifying content vs code strings
+
+## Related Skills
+
+- **contentrain-serve** — Browser-based review UI for extraction approval
+- **contentrain-generate** — SDK client regeneration (required after reuse)
+- **contentrain-sdk** — @contentrain/query usage in source code
+- **contentrain** — Core architecture and full MCP tool catalog

package/skills/contentrain-normalize/references/extraction.md

@@ -113,7 +113,7 @@ Returns a preview of what will be created in `.contentrain/` without making chan
 contentrain_apply(mode: "extract", dry_run: false)
 ```
 
-Creates model definitions and content files in `.contentrain/` on a `contentrain/normalize/extract/{timestamp}` branch. `dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute.
+Creates model definitions and content files in `.contentrain/` on a `cr/normalize/extract/{timestamp}` branch. `dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute.
 
 ### 7. Validate and Submit
 

package/skills/contentrain-normalize/references/reuse.md

@@ -42,7 +42,7 @@ Review the dry-run output:
 contentrain_apply(mode: "reuse", scope: { model: "<model-id>" }, patches: [...], dry_run: false)
 ```
 
-`dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute. This patches source files and creates a `contentrain/normalize/reuse/{model}/{timestamp}` branch.
+`dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute. This patches source files and creates a `cr/normalize/reuse/{model}/{timestamp}` branch.
 
 ### 5. Validate and Submit
 

package/skills/contentrain-quality/SKILL.md

@@ -178,3 +178,10 @@ Noun or noun phrase, sentence case, no trailing colons. Helper text: one sentenc
 - [SEO Rules](references/seo.md) -- SEO optimization rules for content fields, meta tags, and structured data
 - [Accessibility](references/accessibility.md) -- Accessibility rules for content: alt text, ARIA, color contrast, reading level
 - [Media Rules](references/media.md) -- Media asset rules: image optimization, video handling, file naming conventions
+
+## Related Skills
+
+- **contentrain-review** — Apply quality checks during content review
+- **contentrain-content** — Content creation following quality rules
+- **contentrain-translate** — Translation quality and i18n rules
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-quality/references/content-quality.md

@@ -74,6 +74,8 @@ If the agent cannot produce real content, it must leave the entry in `draft` sta
 - Descriptions/excerpts must be unique across all entries in the same collection.
 - If content is shared across entries, extract it to a referenced entry and use a `relation` field.
 - Before writing, query existing entries to confirm no duplication.
+- **Dictionary values:** Different keys must not map to identical values. If "Cancel" exists under `dialog.cancel`, do not create `form.cancel` with the same value — reuse the existing key. MCP warns automatically via `advisories` in the save response.
+- **Cross-model deduplication:** If the same text appears across multiple dictionary models, consider moving it to a shared dictionary or using vocabulary terms for consistency.
 
 ---
 

package/skills/contentrain-review/SKILL.md

@@ -168,3 +168,9 @@ Only make changes after user approval.
 If fixes were applied, call `contentrain_validate` again to confirm all issues are resolved.
 
 Report the final status: "All checks passed" or list remaining issues that require manual attention.
+
+## Related Skills
+
+- **contentrain-serve** — Visual review at http://localhost:3333/content
+- **contentrain-quality** — Quality checklist reference
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-sdk/SKILL.md

@@ -43,7 +43,9 @@ const doc = document('blog-post').bySlug('getting-started')
 |--------|-------------|
 | `.all()` | Get all entries |
 | `.first()` | Get first entry |
-| `.where(field, op, value)` | Filter (`eq`, `ne`, `gt`, `gte`, `lt`, `lte`, `in`, `contains`) |
+| `.count()` | Return count of matching entries |
+| `.where(field, value)` | Equality filter (shorthand for `eq`) |
+| `.where(field, op, value)` | Operator filter: `eq`, `ne`, `gt`, `gte`, `lt`, `lte`, `in`, `contains` |
 | `.sort(field, direction?)` | Sort (`asc`/`desc`, default `asc`) |
 | `.limit(n)` | Limit results |
 | `.offset(n)` | Skip results |
@@ -71,7 +73,9 @@ const doc = document('blog-post').bySlug('getting-started')
 | `.all()` | Get all documents |
 | `.bySlug(slug)` | Get document by slug |
 | `.first()` | Get first document |
-| `.where(field, op, value)` | Filter by frontmatter field |
+| `.count()` | Return count of matching documents |
+| `.where(field, value)` | Equality filter (shorthand) |
+| `.where(field, op, value)` | Operator filter (same ops as QueryBuilder) |
 | `.include(relation)` | Resolve relation fields |
 
 ## CDN Mode (Remote Data)
@@ -107,6 +111,28 @@ const doc = await client.document('docs').locale('en').bySlug('intro')
 | `in` | `.where('category', 'in', ['a','b'])` | In array |
 | `contains` | `.where('tags', 'contains', 'vue')` | String/array contains |
 
+### CDN Entry Metadata, Media & Forms
+
+```typescript
+// Entry metadata (status, publish_at, expire_at)
+const posts = await client.collection('blog').locale('en').withMeta().all()
+
+// Media manifest & variant URLs
+const media = client.media()
+const asset = await media.asset('hero.jpg')
+const url = media.url(asset, 'thumb')
+
+// Forms (config fetch + submit)
+const form = client.form()
+const config = await form.config('contact')
+const result = await form.submit('contact', { name: 'Alice' })
+
+// Conversation API (external AI content operations)
+const conv = client.conversation()
+const response = await conv.send('Create a blog post about Vue 4')
+const history = await conv.history(response.conversationId)
+```
+
 ### CDN vs Local
 
 | Aspect | Local (`#contentrain`) | CDN (`createContentrain()`) |
@@ -143,3 +169,9 @@ const doc = await client.document('docs').locale('en').bySlug('intro')
 | Reference | Description |
 |-----------|-------------|
 | [Bundler Configuration](references/bundler-config.md) | Vite, Next.js, Nuxt, SvelteKit, Metro alias setup |
+
+## Related Skills
+
+- **contentrain-generate** — Generate the SDK client before using it
+- **contentrain-quality** — Content quality rules for entries you query
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-serve/SKILL.md

@@ -94,3 +94,10 @@ The serve UI is a **monitoring + approval surface**, not an action trigger:
 - **Prompts**: every page shows copyable agent prompts that developers paste into their AI agent
 
 All mutations (create, edit, delete, scan, fix, normalize) are **agent-driven via MCP tools**. The UI never triggers these directly.
+
+## Related Skills
+
+- **contentrain-normalize** — Normalize workflow uses serve for extraction/reuse review and approval
+- **contentrain-diff** — Terminal-based branch review alternative to serve /branches page
+- **contentrain-review** — Content review workflow uses serve for visual inspection
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-translate/SKILL.md

@@ -135,6 +135,14 @@ After all translations are saved, re-generate the SDK client so the application
 npx contentrain generate
 ```
 
+### After Translation: SDK Regeneration
+
+New locale data requires SDK client regeneration:
+
+```bash
+npx contentrain generate
+```
+
 ### 8. Save Translations
 
 After approval, call `contentrain_content_save` for each model and target locale:
@@ -162,7 +170,7 @@ If validation fails, fix issues and re-save.
 
 Call `contentrain_submit` to commit the translations:
 
-- Branch: `contentrain/content/{model}/{targetLocale}/{timestamp}`.
+- Branch: `cr/content/{model}/{targetLocale}/{timestamp}`.
 - Each locale can be submitted independently.
 
 ### 11. Final Summary
@@ -178,3 +186,10 @@ Report to the user:
 - Submission status.
 - Reminder to review translations for nuance that automated checks cannot catch.
 - Suggestion: run `/contentrain-review` to verify overall i18n coverage.
+
+## Related Skills
+
+- **contentrain-serve** — Review translated content in browser
+- **contentrain-generate** — SDK regeneration for new locale data
+- **contentrain-quality** — Translation quality rules
+- **contentrain** — Core architecture and MCP tool catalog

package/skills/contentrain-validate-fix/SKILL.md

@@ -90,3 +90,9 @@ If validation still fails:
 
 - summarize remaining blockers precisely
 - tell the user which model/entry needs manual attention
+
+## Related Skills
+
+- **contentrain-review** — Full content review workflow
+- **contentrain-content** — Fix content entries after validation errors
+- **contentrain** — Core architecture and MCP tool catalog

package/workflows/contentrain-normalize.md

@@ -134,7 +134,7 @@ Do NOT proceed to apply unless the user approves (either via UI or explicit conf
 
 4. **Detect the user's decision.** After directing the user to the UI, check the plan file to determine the outcome:
    - **Plan file exists** (`status: "pending"`) → user has not decided yet — wait
-   - **Plan file deleted + new `contentrain/normalize/extract/*` branch exists** → user approved, extraction applied
+   - **Plan file deleted + new `cr/normalize/extract/*` branch exists** → user approved, extraction applied
    - **Plan file deleted + no new branch** → user rejected the plan
 
    If rejected, ask the user what to change and iterate from Step 4 (re-evaluate candidates).
@@ -161,7 +161,7 @@ After user approval (via UI or chat), call `contentrain_apply(mode: "extract", d
 
 Note: `dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute.
 
-This creates model definitions and content files in `.contentrain/` on a `contentrain/normalize/extract/{timestamp}` branch. Source files are NOT modified.
+This creates model definitions and content files in `.contentrain/` on a `cr/normalize/extract/{timestamp}-{suffix}` branch. Source files are NOT modified.
 
 If approved via UI, the UI calls this automatically — no additional agent action needed.
 
@@ -230,7 +230,7 @@ After user confirmation, call `contentrain_apply(mode: "reuse", scope: { model:
 
 Note: `dry_run` defaults to `true`, so you MUST explicitly set `dry_run: false` to execute.
 
-This patches source files and creates a `contentrain/normalize/reuse/{model}/{timestamp}` branch.
+This patches source files and creates a `cr/normalize/reuse/{model}/{timestamp}-{suffix}` branch.
 
 ### Step 5. Validate and Submit
 

package/workflows/contentrain-translate.md

@@ -154,7 +154,7 @@ If validation fails, fix issues and re-save.
 
 Call `contentrain_submit` to commit the translations:
 
-- Branch: `contentrain/content/{model}/{targetLocale}/{timestamp}`.
+- Branch: `cr/content/{model}/{targetLocale}/{timestamp}-{suffix}`.
 - Each locale can be submitted independently.
 
 ### 11. Final Summary