@ekkos/cli 0.2.0

package/templates/claude-plugins/blog-manager/commands/blog.md

@@ -0,0 +1,691 @@
+# Blog Manager
+
+Editorial enforcement system for ekkOS blog. Create, research, and publish thought leadership content with built-in credibility guardrails.
+
+## Overview
+
+**This is not a blog generator. It is an institutional credibility engine.**
+
+The ekkOS blog exists to educate, not to sell. Every post must:
+- Teach first, mention ekkOS last
+- Ground claims in evidence or explicit qualifiers
+- Acknowledge trade-offs and limitations
+- Position ekkOS as one option, not the only option
+
+## Editorial Philosophy: "Teach First, Sell Last"
+
+**Content Ratio:** 80% problem + solutions landscape, 15% practical how-to, 5% ekkOS mention
+
+**The Rule:** ekkOS should only appear in a dedicated section titled "How we think about this at ekkOS_" — 3-6 sentences max, placed near the end.
+
+## Usage
+
+```bash
+# Dashboard - see all posts and their status
+/blog
+
+# Create new post with guided workflow
+/blog new [topic]
+
+# Edit existing post
+/blog edit [slug]
+
+# Preview post with frontmatter validation
+/blog preview [slug]
+
+# Generate cover image with FLUX 2 Max
+/blog image [slug]
+
+# Schedule post for future publication
+/blog schedule [slug] [date]
+
+# Publish immediately
+/blog publish [slug]
+
+# SEO optimization (problem-oriented, not product-oriented)
+/blog seo [slug]
+
+# Research trending topics with Perplexity Sonar Pro
+/blog research [topic]
+
+# Generate topic ideas relevant to ekkOS mission
+/blog ideas
+
+# Editorial QA - salesiness score, banned words, truthfulness
+/blog tonecheck [slug]
+
+# Extract and audit all claims
+/blog claims [slug]
+```
+
+---
+
+## Example - Dashboard (`/blog`)
+
+```
+📰 ekkOS Blog Dashboard
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📊 Status Overview:
+   Published: 7 posts
+   Scheduled: 0 posts
+   Drafts: 0 posts
+
+───────────────────────────────────────────────────────────────────
+✅ Published Posts (newest first):
+
+1. patterns-that-learn.md
+   "Patterns That Learn — How AI Memory Should Actually Work"
+   📅 2026-01-14 · 📖 5 min · 🏷️ patterns, feedback-loops, learning
+
+2. instruction-hierarchy-problem.md
+   "The Instruction Hierarchy Problem in AI Safety"
+   📅 2026-01-12 · 📖 6 min · 🏷️ safety, directives, architecture
+
+3. one-memory-five-tools.md
+   "One Memory, Five Tools — Why Platform Matters"
+   📅 2026-01-10 · 📖 5 min · 🏷️ integration, tools, platform
+
+4. why-jailbreaks-work-and-how-persistent-memory-fixes-them.md
+   "Why Jailbreaks Work (And How Persistent Memory Fixes Them)"
+   📅 2026-01-08 · 📖 4 min · 🏷️ security, jailbreaks, memory
+
+5. your-ai-forgot-again.md
+   "Your AI Forgot Again — The Context Window Crisis"
+   📅 2026-01-06 · 📖 4 min · 🏷️ context, memory, limitations
+
+6. why-rag-isnt-memory.md
+   "Why RAG Isn't Memory"
+   📅 2026-01-04 · 📖 5 min · 🏷️ RAG, memory, architecture
+
+7. welcome-to-ekkos-blog.md
+   "Welcome to the ekkOS Blog"
+   📅 2026-01-02 · 📖 2 min · 🏷️ announcement
+
+───────────────────────────────────────────────────────────────────
+Commands:
+  /blog new [topic]     Create new post
+  /blog edit [slug]     Edit existing post
+  /blog tonecheck [slug] Run editorial QA
+  /blog research [topic] Research with Perplexity
+```
+
+---
+
+## Example - New Post (`/blog new`)
+
+```
+/blog new "Why Most AI Memory Solutions Miss the Point"
+
+📝 New Post Workflow
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Topic: "Why Most AI Memory Solutions Miss the Point"
+
+───────────────────────────────────────────────────────────────────
+📋 Step 1: Choose Template
+
+Which template fits this topic?
+
+[A] Pain Point Deep Dive (recommended)
+    • Why this problem exists (systems + incentives)
+    • Where teams usually get misled
+    • Failure modes and measurement
+    • 2-3 approaches with trade-offs
+    • "Where ekkOS_ fits (and where it doesn't)"
+    • Practical next steps checklist
+
+[B] Field Guide
+    • Definitions
+    • Decision tree
+    • Implementation pitfalls
+    • Minimal reference architecture
+    • Validation plan
+    • Optional ekkOS_ sidebar
+
+[C] Research Commentary
+    • What the source says (quoted minimally, cited)
+    • What it gets right
+    • What it misses in deployment reality
+    • Your framework or implementation interpretation
+    • ekkOS_ positioning as practical path
+
+> A
+
+───────────────────────────────────────────────────────────────────
+📋 Step 2: Evidence vs Hypothesis
+
+What can we claim WITH evidence?
+(Benchmarks, citations, public docs, internal metrics with scope)
+
+> Vector DBs don't track outcome success rates (cite LangChain docs)
+> RAG systems retrieve but don't learn (cite industry surveys)
+> Pattern-based retrieval improves over time (internal: 84% success rate)
+
+What must be framed as HYPOTHESIS or EXPERIENCE?
+
+> "In our deployments, we've seen..." (scope: 50+ users)
+> "Teams often struggle with..." (experience, not universal)
+
+───────────────────────────────────────────────────────────────────
+📋 Step 3: Trade-offs
+
+What are the top 3 trade-offs of ekkOS's approach?
+
+> 1. Requires user adoption of pattern forging habits
+> 2. Cold start problem - needs initial patterns to be useful
+> 3. Verification adds latency vs pure retrieval
+
+───────────────────────────────────────────────────────────────────
+📋 Step 4: Where We Don't Fit
+
+Write the "where ekkOS_ doesn't fit" paragraph:
+
+> ekkOS_ is not the right choice if you need: (a) pure vector search
+> without learning, (b) one-shot retrieval without feedback loops,
+> or (c) systems where patterns don't repeat across sessions.
+
+───────────────────────────────────────────────────────────────────
+📋 Step 5: Generate Metadata
+
+Title: "Why Most AI Memory Solutions Miss the Point"
+Slug: why-most-ai-memory-solutions-miss-the-point
+Description: "Vector databases store context. RAG retrieves it.
+But neither learns from outcomes. Here's what's missing."
+Tags: ["memory", "RAG", "vector-db", "learning", "architecture"]
+Author: ekkOS Team
+
+───────────────────────────────────────────────────────────────────
+✅ Creating post...
+
+Created: apps/blog/content/why-most-ai-memory-solutions-miss-the-point.md
+
+📝 Post created as DRAFT. Next steps:
+   /blog edit why-most-ai-memory-solutions-miss-the-point
+   /blog tonecheck why-most-ai-memory-solutions-miss-the-point
+   /blog image why-most-ai-memory-solutions-miss-the-point
+   /blog publish why-most-ai-memory-solutions-miss-the-point
+```
+
+---
+
+## Example - Tone Check (`/blog tonecheck`)
+
+```
+/blog tonecheck patterns-that-learn
+
+📝 Tone Check: patterns-that-learn.md
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📊 Salesiness Score: 8% ✅ (threshold: <15%)
+
+   Metrics:
+   • ekkOS mentions: 3 (all in final section) ✅
+   • First mention: line 142 (87% into post) ✅
+   • Product density: 2.1% ✅
+   • Banned words found: 0 ✅
+   • "Simply/Just" found: 0 ✅
+
+───────────────────────────────────────────────────────────────────
+✅ Truthfulness Checklist:
+
+   [✓] Numeric claims sourced
+       • "60-70% drop" → The Hacker News (cited line 38)
+
+   [✓] Comparisons qualified
+       • "in many setups" used appropriately
+
+   [✓] No unqualified roadmap claims
+
+   [✓] Has "Trade-offs" section (line 98)
+
+───────────────────────────────────────────────────────────────────
+⚠️ Warnings (advisory, not blocking):
+
+   • Line 45: "often fails" → Consider adding scope qualifier
+   • Consider adding explicit "Limitations" subsection
+
+───────────────────────────────────────────────────────────────────
+🚫 Blockers: None
+
+Overall: PASS ✅
+
+Ready to publish? Run: /blog publish patterns-that-learn
+```
+
+---
+
+## Example - Claims Audit (`/blog claims`)
+
+```
+/blog claims patterns-that-learn
+
+📋 Claims Audit: patterns-that-learn.md
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Found 7 claims (declarative, numeric, or comparative):
+
+───────────────────────────────────────────────────────────────────
+1. "False positives drop by 60-70% in first feedback cycle"
+
+   Type: OBSERVED ✅
+   Source: The Hacker News (cited line 38) ✅
+   Status: Valid
+
+───────────────────────────────────────────────────────────────────
+2. "Most knowledge bases don't track which solutions work"
+
+   Type: EXPERIENCE ⚠️
+   Qualifier: Missing scope
+
+   Suggested fix:
+   "In our analysis of 20+ knowledge base implementations,
+   most don't track which solutions actually work."
+
+───────────────────────────────────────────────────────────────────
+3. "ekkOS patterns surface more effective solutions over time"
+
+   Type: COMPARATIVE ⚠️
+   Issue: Needs conditions defined
+
+   Suggested fix:
+   "In systems with active feedback loops, ekkOS patterns
+   surface more effective solutions over time, as measured
+   by success rate improvements in our internal deployments."
+
+───────────────────────────────────────────────────────────────────
+4. "Pattern success rates improve with feedback"
+
+   Type: OBSERVED ✅
+   Source: Internal metrics (labeled line 112) ✅
+   Status: Valid
+
+───────────────────────────────────────────────────────────────────
+5. "Static knowledge bases don't learn"
+
+   Type: OBSERVED ✅
+   Source: Architectural definition (self-evident) ✅
+   Status: Valid
+
+───────────────────────────────────────────────────────────────────
+6. "The principle applies directly to development knowledge"
+
+   Type: EXPERIENCE ✅
+   Qualifier: "The principle" (referential, scoped) ✅
+   Status: Valid
+
+───────────────────────────────────────────────────────────────────
+7. "Without this, your knowledge base is just a filing cabinet"
+
+   Type: COMPARATIVE ✅
+   Qualifier: Metaphor, clearly rhetorical ✅
+   Status: Valid
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Summary:
+   ✅ Valid: 5 claims
+   ⚠️ Needs qualifier: 2 claims
+   🚫 Invalid: 0 claims
+
+Actions Required:
+   [ ] Add scope to claim #2
+   [ ] Define conditions for claim #3
+```
+
+---
+
+## Example - Research (`/blog research`)
+
+```
+/blog research "AI agent memory systems January 2026"
+
+📡 Researching with Perplexity Sonar Pro...
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+🔍 Query: "AI agent memory systems developments January 2026"
+📅 Recency: Last 30 days
+
+───────────────────────────────────────────────────────────────────
+📰 Key Findings:
+
+1. **OpenAI Agents SDK Update** (Jan 5, 2026)
+   Native memory persistence announced for multi-turn agents
+   Source: [OpenAI Blog](https://openai.com/blog/agents-sdk-memory)
+
+   Gap: Still retrieval-only, no verified learning loop
+
+2. **LangChain Memory Module v0.3** (Jan 3, 2026)
+   New ConversationBufferMemory with persistence
+   Source: [LangChain Docs](https://docs.langchain.com/memory)
+
+   Gap: Storage without outcome tracking
+
+3. **Stanford HAI Paper: "Memory-Augmented Agents"** (Dec 28, 2025)
+   Research on episodic memory for LLM agents
+   Source: [arXiv](https://arxiv.org/abs/2512.xxxxx)
+
+   Insight: Validates feedback-driven pattern evolution
+
+4. **Hacker News Discussion: "RAG is Not Memory"** (Jan 2, 2026)
+   200+ comments on memory vs retrieval distinction
+   Source: [HN Thread](https://news.ycombinator.com/item?id=xxxxx)
+
+   Angle: Community already understands the problem
+
+───────────────────────────────────────────────────────────────────
+💡 ekkOS Blog Angles (aligned with editorial stance):
+
+1. "What OpenAI's Agents SDK Gets Right About Memory (And What's Missing)"
+   Template: Research Commentary
+   Angle: Acknowledge progress, identify verification gap
+   ekkOS fit: Where feedback loops matter
+
+2. "The Stanford HAI Paper Validates What We've Been Building"
+   Template: Research Commentary
+   Angle: Academic support for pattern evolution
+   ekkOS fit: Implementation of their research direction
+
+3. "Why the HN Community is Right: RAG Really Isn't Memory"
+   Template: Pain Point Deep Dive
+   Angle: Expand on community insight with architecture depth
+   ekkOS fit: One approach to the problem they identified
+
+───────────────────────────────────────────────────────────────────
+📋 Claim Validation Opportunities:
+
+Before writing, validate with Perplexity:
+• "What benchmarks exist for agent memory systems?"
+• "Criticisms of RAG-only approaches in production"
+• "Failure modes of LangChain memory in long sessions"
+
+Would you like to create a post on any of these topics?
+```
+
+---
+
+## Example - Topic Ideas (`/blog ideas`)
+
+```
+/blog ideas
+
+💡 Topic Ideas for ekkOS Blog
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Based on: ekkOS mission, current trends, content gaps
+
+───────────────────────────────────────────────────────────────────
+🔥 High Priority (trending + aligned):
+
+1. "The Verification Gap in AI Agent Memory"
+   Why: OpenAI/LangChain memory lacks outcome tracking
+   Template: Pain Point Deep Dive
+   Evidence available: Yes (public docs, benchmarks)
+
+2. "How Feedback Loops Change AI Knowledge Management"
+   Why: Academic research supports this direction
+   Template: Research Commentary
+   Evidence available: Yes (Stanford HAI paper)
+
+3. "What Enterprise Teams Get Wrong About RAG"
+   Why: Perplexity shows enterprise RAG struggles
+   Template: Field Guide
+   Evidence available: Yes (industry surveys)
+
+───────────────────────────────────────────────────────────────────
+📚 Educational (evergreen):
+
+4. "A Field Guide to AI Memory Architectures"
+   Why: Foundational content, drives organic search
+   Template: Field Guide
+   Evidence available: Yes (architectural definitions)
+
+5. "Measuring AI Agent Improvement Over Time"
+   Why: Practical how-to, unique to learning systems
+   Template: Field Guide
+   Evidence available: Internal metrics + methodology
+
+───────────────────────────────────────────────────────────────────
+⚠️ Avoid (too promotional or unsubstantiated):
+
+• "Why ekkOS is Better Than LangChain" (comparative without conditions)
+• "The Future of AI Memory" (aspirational, no evidence)
+• "ekkOS Success Stories" (case study needs real customers)
+
+───────────────────────────────────────────────────────────────────
+Ready to write? Run: /blog new "[topic]"
+```
+
+---
+
+## Example - SEO Optimization (`/blog seo`)
+
+```
+/blog seo patterns-that-learn
+
+🔍 SEO Optimization: patterns-that-learn.md
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Current metadata:
+   Title: "Patterns That Learn — How AI Memory Should Actually Work"
+   Description: "Static knowledge bases don't improve. Feedback
+   loops do. Here's how patterns can evolve based on whether they
+   actually solve problems."
+
+───────────────────────────────────────────────────────────────────
+✅ SEO Compliance Check:
+
+   [✓] Title is problem-oriented (not product-oriented)
+   [✓] Description promises learning (not product outcomes)
+   [✓] Tags are topical: patterns, feedback-loops, learning, architecture
+   [✓] No brand-heavy tags
+
+───────────────────────────────────────────────────────────────────
+💡 Optimization Suggestions:
+
+   Title alternatives (A/B test candidates):
+   • "Why Static Knowledge Bases Don't Learn (And What Does)"
+   • "Feedback Loops for AI Knowledge: A Practical Guide"
+   • "The Missing Piece in AI Knowledge Management"
+
+   Description alternatives:
+   • "Most knowledge bases just store information. They don't know
+     which solutions actually work. Here's how to fix that with
+     feedback-driven patterns."
+
+   Additional tags to consider:
+   • knowledge-management
+   • developer-tools
+   • ai-infrastructure
+
+───────────────────────────────────────────────────────────────────
+🚫 SEO Anti-patterns (avoided):
+
+   ✓ Not using: "ekkOS" in title
+   ✓ Not using: "best", "ultimate", "revolutionary"
+   ✓ Not promising: product outcomes in description
+
+Update metadata? (yes/no)
+```
+
+---
+
+## Frontmatter Reference
+
+All posts use this schema:
+
+```yaml
+---
+title: "Post Title Here"
+description: "SEO description under 160 characters. Promise learning, not product outcomes."
+date: "2026-01-14T09:00:00-05:00"  # ISO 8601, controls publication
+author: "ekkOS Team"
+tags: ["tag1", "tag2", "tag3"]  # Topical, not brand-heavy
+image: "/images/blog/slug-name.png"
+imageAlt: "Descriptive alt text for accessibility"
+draft: true  # Optional - hides from public until removed
+---
+```
+
+**Publication Logic:**
+- `draft: true` → Hidden from public, visible at /preview
+- `date > now` → Scheduled, hidden until date passes
+- `draft: false && date <= now` → Published
+
+---
+
+## Banned Words and Constructions
+
+**Hard ban (will fail tonecheck):**
+- revolutionary
+- game-changing
+- unparalleled
+- best-in-class
+- world-class
+- ultimate
+
+**Soft ban (warnings):**
+- "Simply..." (minimizes complexity)
+- "Just..." (minimizes complexity)
+- "Our solution is better than..." (unqualified comparative)
+
+**Preferred language:**
+- trade-offs
+- failure modes
+- constraints
+- what to measure
+- how to validate
+- in practice
+- in many setups
+- depends on
+
+---
+
+## Claim Categories
+
+Every claim in a post should fall into one of these categories:
+
+| Category | Definition | Requirement |
+|----------|------------|-------------|
+| **Observed** | Backed by citation, benchmark, public doc | Must cite source |
+| **Experience** | "We've seen..." or "In our deployments..." | Must include scope |
+| **Aspirational** | Planned features or roadmap | Must label "planned" or "in progress" |
+| **Comparative** | "X is better than Y" | Must define conditions |
+
+**Rule:** If a paragraph includes a product claim, it must include either (a) a citation, (b) a concrete constraint, or (c) an explicit "depends on" qualifier.
+
+---
+
+## Competitor Mentions
+
+When mentioning other tools (LangChain, LlamaIndex, etc.):
+
+**Do:**
+- Critique approaches, not teams
+- Use "in many setups" and "often" rather than absolutes
+- Acknowledge what they do well
+- Identify gaps that appear at scale
+
+**Don't:**
+- Assert intent or competence of another team
+- Make unqualified "better than" claims
+- Use dismissive language
+
+**Pattern:**
+"Here's what [tool] is strong at; here's the gap that shows up at scale; here's one way to address it."
+
+---
+
+## The "ekkOS_ Promo with Class" Formula
+
+Every post gets ONE small section with this structure:
+
+**Section title:** "How we think about this at ekkOS_"
+
+**Contents (3-6 sentences max):**
+1. One concrete capability
+2. One explicit constraint or trade-off
+3. One "how to evaluate" suggestion
+
+**Example:**
+> ### How we think about this at ekkOS_
+>
+> We address this by tracking outcome success rates for every pattern retrieved.
+> When a pattern helps solve a problem, its weight increases; when it doesn't, it
+> decreases. This requires users to close the feedback loop, which adds friction
+> but enables genuine learning. If you're evaluating memory systems, ask: "Does
+> this system know which of its suggestions actually worked?"
+
+---
+
+## Implementation Details
+
+When this command runs, Claude will:
+
+**For `/blog` (dashboard):**
+1. Use `Glob` to find all `*.md` files in `apps/blog/content/`
+2. Use `Read` to parse frontmatter from each file
+3. Categorize by status (published/scheduled/draft)
+4. Display sorted by date
+
+**For `/blog new`:**
+1. Guide through template selection
+2. Ask evidence vs hypothesis questions
+3. Ask trade-offs questions
+4. Ask "where we don't fit" question
+5. Generate slug from title
+6. Use `Write` to create markdown file with frontmatter + template structure
+
+**For `/blog tonecheck`:**
+1. Use `Read` to get post content
+2. Count ekkOS mentions and their positions
+3. Check for banned words (regex match)
+4. Verify truthfulness checklist items exist
+5. Calculate salesiness score
+6. Output advisory warnings + any blockers
+
+**For `/blog claims`:**
+1. Use `Read` to get post content
+2. Extract declarative sentences, numeric claims, comparative phrases
+3. Categorize each claim (Observed/Experience/Aspirational/Comparative)
+4. Check for required citations or qualifiers
+5. Output checklist with suggested fixes
+
+**For `/blog research`:**
+1. Call `mcp__perplexity__search` or `mcp__perplexity__reason` with topic
+2. Parse results for trends, citations, gaps
+3. Map findings to ekkOS blog angles
+4. Suggest claim validation queries
+
+**For `/blog image`:**
+1. Read post to understand topic
+2. Generate FLUX 2 Max prompt following blog image style
+3. Run `apps/blog/scripts/generate-blog-images.mjs` or generate inline
+4. Save to `apps/blog/public/images/blog/`
+
+---
+
+## Requirements
+
+- ekkOS blog infrastructure in `apps/blog/`
+- Perplexity MCP server configured (for research commands)
+- FLUX 2 Max via Vercel AI Gateway (for image generation)
+- File system access via Claude Code tools
+
+---
+
+## Content Location
+
+- Posts: `apps/blog/content/*.md`
+- Images: `apps/blog/public/images/blog/`
+- Preview: Visit `/preview` route in running blog
+
+---
+
+## Quality Standards
+
+This is an editorial enforcement system. The goal is thought leadership that builds credibility, not marketing copy that generates skepticism.
+
+**The test:** Would a senior engineer at a competitor read this and think "that's a fair and useful analysis" rather than "that's just marketing"?
+
+If yes, publish. If no, revise.

package/templates/claude-plugins/golden-loop-monitor/.claude-plugin/plugin.json

@@ -0,0 +1,8 @@
+{
+  "name": "ekkos-golden-loop-monitor",
+  "description": "Monitor and troubleshoot the Golden Loop (RETRIEVE → APPLY → MEASURE → LEARN). Ensure ekkOS is working correctly for your codebase.",
+  "author": {
+    "name": "ekkOS",
+    "email": "support@ekkos.dev"
+  }
+}