# Cortex - Claude's Cognitive Layer ## Metadata name: cortex description: Cross-session memory with deep reasoning - query, reflect, learn, consolidate version: 2.0.0 triggers: - /cortex - memory - remember - what did I do - past sessions - learnings ## Overview Cortex provides persistent memory across Claude Code sessions. This skill exposes the Cortex MCP tools as user-friendly commands with visual feedback. ## Commands | Command | Description | Cost | |---------|-------------|------| | `/cortex` | Show status and available commands | Free | | `/cortex help` | Show detailed help with examples | Free | | `/cortex query ` | Search stored memories | ~$0.001 | | `/cortex recall` | Get context-aware memories for current task | ~$0.001 | | `/cortex reflect ` | Deep meta-cognitive analysis | ~$0.01 | | `/cortex infer ` | Find connections between ideas | ~$0.01 | | `/cortex learn ` | Store a new learning | ~$0.01 | | `/cortex consolidate` | Clean up and organize memories | ~$0.02 | | `/cortex stats` | Show memory counts and API costs | Free | | `/cortex health` | Verify all systems are working | Free | | `/cortex log` | View recent activity log | Free | | `/cortex config` | View/edit configuration | Free | | `/cortex export [format]` | Export memories (json/md) | Free | | `/cortex session` | Show current session summary | Free | ## Visual Feedback When Cortex processes, display neural activity: ``` ╭──────────────────────────────────╮ │ 🧠 CORTEX ══════════════════════│ │ · ∿ · ∿ · ∿ · │ │ ◇━━━◇━━━◇ ◆━━━◆━━━◆ │ │ Haiku Sonnet │ │ querying... │ ╰──────────────────────────────────╯ ``` ## Instructions When the user invokes `/cortex` or asks about memories/past sessions: ### 0. Help Command (`/cortex help`) Display comprehensive help with groupings: ``` ╭─────────────────────────────────────────────────╮ │ 🧠 CORTEX v2.0 - Claude's Memory Layer │ ├─────────────────────────────────────────────────┤ │ QUICK │ │ /cortex Status overview │ │ /cortex help This screen │ │ /cortex stats Memory counts & costs │ │ /cortex health Verify everything works │ │ │ │ MEMORY (Haiku - fast, ~$0.001) │ │ /cortex query "X" Search memories │ │ /cortex recall Context retrieval │ │ │ │ REASONING (Sonnet - deep, ~$0.01) │ │ /cortex reflect "X" Meta-analysis │ │ /cortex infer A B Find connections │ │ /cortex learn "X" Store insight │ │ /cortex consolidate Clean up │ │ │ │ UTILITIES │ │ /cortex log Recent activity │ │ /cortex config View/edit settings │ │ /cortex export Export memories │ │ /cortex session Current session info │ │ │ │ NATURAL LANGUAGE │ │ "What did I do last time?" │ │ "Remember: always validate JWT first" │ │ "What patterns am I seeing?" │ │ "How does X connect to Y?" │ ╰─────────────────────────────────────────────────╯ ``` ### 1. Status Command (`/cortex` or `/cortex status`) Display current Cortex status: ``` ╭─────────────────────────────────────────────╮ │ 🧠 CORTEX STATUS │ ├─────────────────────────────────────────────┤ │ MCP Server: ✓ Connected │ │ Memories: 127 total │ │ • Skills: 23 │ │ • Patterns: 34 │ │ • Learnings: 45 │ │ • Insights: 25 │ │ Last Activity: 2 hours ago │ │ Session Cost: $0.02 │ ╰─────────────────────────────────────────────╯ ``` To get this data: 1. Count lines in `~/.claude/memory/data/memories/*.jsonl` 2. Check `~/.claude/memory/logs/` for last activity 3. Read stats from MCP server if available ### 2. Query Command (`/cortex query `) ``` ╭──────────────────────────────────────╮ │ 🧠 CORTEX ═══════════════════════ │ │ · ∿ · ∿ · │ │ ◇━━━◇━━━◇ Haiku searching... │ ╰──────────────────────────────────────╯ ``` Use the `cortex__query` MCP tool with the search term. Display results with relevance scores. ### 3. Reflect Command (`/cortex reflect `) ``` ╭──────────────────────────────────────╮ │ 🧠 CORTEX ═══════════════════════ │ │ · ∿ · ∿ · │ │ ◆━━━◆━━━◆ Sonnet thinking...│ ╰──────────────────────────────────────╯ ``` Use the `cortex__reflect` MCP tool. This uses Sonnet for deep reasoning. Include depth parameter: quick (512 tokens), moderate (1024), deep (2048). ### 4. Learn Command (`/cortex learn `) ``` ╭──────────────────────────────────────╮ │ 🧠 CORTEX ═══════════════════════ │ │ · ∿ · ∿ · │ │ ◆━━━◆━━━◆ Sonnet analyzing..│ ╰──────────────────────────────────────╯ ``` Use the `cortex__learn` MCP tool to store the insight. Report quality score and whether it was stored. ### 5. Stats Command (`/cortex stats`) Read and display: - Memory file sizes and counts - API token usage from server stats - Estimated costs (Haiku: $0.25/1M, Sonnet: $3/1M) ``` ╭─────────────────────────────────────────────╮ │ 🧠 CORTEX STATS │ ├─────────────────────────────────────────────┤ │ MEMORIES │ │ Skills: 23 entries (12 KB) │ │ Patterns: 34 entries (8 KB) │ │ Learnings: 45 entries (15 KB) │ │ Insights: 25 entries (6 KB) │ │ ───────────────────────────────── │ │ Total: 127 entries (41 KB) │ ├─────────────────────────────────────────────┤ │ API USAGE (This Session) │ │ Haiku: 1,234 tokens ~$0.0003 │ │ Sonnet: 5,678 tokens ~$0.017 │ │ ───────────────────────────────── │ │ Total Cost: ~$0.02 │ ╰─────────────────────────────────────────────╯ ``` ### 6. Health Command (`/cortex health`) Run comprehensive health check: ``` ╭─────────────────────────────────────────────────╮ │ 🧠 CORTEX HEALTH CHECK │ ├─────────────────────────────────────────────────┤ │ MCP Server ✓ Connected │ │ Memory Files ✓ 5 files (127 entries) │ │ API Key ✓ Valid (sk-ant-...) │ │ Hooks ✓ SessionStart, SessionEnd │ │ Last Activity 2 hours ago │ │ Disk Usage 41 KB │ ├─────────────────────────────────────────────────┤ │ STATUS: All systems operational ✓ │ ╰─────────────────────────────────────────────────╯ ``` To verify: 1. Check MCP connection: `claude mcp list | grep cortex` 2. Check files exist: `ls ~/.claude/memory/data/memories/` 3. Check API key: `[ -n "$ANTHROPIC_API_KEY" ] && echo "Set"` 4. Check hooks: `grep -A3 "SessionStart" ~/.claude/settings.json` ### 7. Log Command (`/cortex log`) Show recent activity from logs: ``` ╭─────────────────────────────────────────────────╮ │ 🧠 CORTEX ACTIVITY LOG (last 10) │ ├─────────────────────────────────────────────────┤ │ 14:32 query "authentication" → 3 hits │ │ 14:28 learn "JWT pattern" → stored │ │ 14:15 recall (session start) → 2 items │ │ 13:45 reflect "debugging approach" → complete│ │ ... │ ╰─────────────────────────────────────────────────╯ ``` Read from: `~/.claude/memory/logs/*.log` ### 8. Config Command (`/cortex config`) Show current configuration: ``` ╭─────────────────────────────────────────────────╮ │ 🧠 CORTEX CONFIGURATION │ ├─────────────────────────────────────────────────┤ │ SESSION START │ │ Max memories injected: 5 │ │ Relevance threshold: 0.3 │ │ │ │ SESSION END │ │ Quality threshold: 0.4 │ │ Max extractions: 10 │ │ │ │ LADS (Learning) │ │ Evolution enabled: yes │ │ Evolution interval: 24h │ │ │ │ Config file: ~/.claude/memory/data/configs/ │ │ current.json │ ╰─────────────────────────────────────────────────╯ ``` Read from: `~/.claude/memory/data/configs/current.json` ### 9. Export Command (`/cortex export [format]`) Export memories to file: ``` /cortex export # Default: markdown /cortex export json # JSON format /cortex export md # Markdown format ``` Output example for markdown: ```markdown # Cortex Memory Export Generated: 2026-01-27 ## Skills (23 entries) - Authentication: Use JWT with refresh tokens... - Error handling: Always log context... ## Learnings (45 entries) - "Always validate JWT expiry before permissions" - "Use circuit breaker for external API calls" ... ``` ### 10. Session Command (`/cortex session`) Show current session summary: ``` ╭─────────────────────────────────────────────────╮ │ 🧠 CORTEX SESSION │ ├─────────────────────────────────────────────────┤ │ Started: 14:00 (2h 32m ago) │ │ Working Dir: ~/.claude/memory │ │ Git Branch: master │ │ │ │ INJECTED AT START │ │ • JWT authentication pattern │ │ • Error handling best practices │ │ │ │ WILL EXTRACT AT END │ │ • MCP Resources implementation approach │ │ • Skill enhancement patterns │ │ │ │ API Usage: ~$0.02 │ ╰─────────────────────────────────────────────────╯ ``` This shows what memories were injected at session start and previews what will be extracted at session end. ## Auto-Activation This skill should auto-activate when the user says: - "What did I do last time?" - "Do you remember when..." - "Check my past sessions" - "What have I learned about..." - "Search my memories for..." When auto-activated, show the visual feedback and use the appropriate tool. ## Error Handling If MCP server is not connected: ``` ╭──────────────────────────────────────╮ │ 🧠 CORTEX ═══════════════════════ │ │ ✗ MCP Server Not Connected │ │ │ │ Try: claude mcp list │ │ Check: ~/.claude.json │ ╰──────────────────────────────────────╯ ``` ## Examples **User**: `/cortex query authentication` **Response**: Shows neural animation, then displays matching memories about authentication patterns. **User**: "What approach did I use for the login system?" **Response**: Auto-activates, queries memories, shows relevant past decisions. **User**: `/cortex learn "Always validate JWT expiry before checking permissions"` **Response**: Analyzes insight quality, stores if good, reports result.