claude-memory/CORE.md

Memory Core (auto-generated)

Last updated: 2026-03-02 | Active memories: 194/518 | Next refresh: daily (systemd timer)

Critical Solutions

• N8N + Claude Code Automated Server Monitoring System - Built automated server health monitoring with N8N and Claude Code headless mode. (n8n, claude-code, automation)
• Claude Scheduled Tasks: headless Claude Code on systemd timers - Built a "Claude Cowork"-style scheduled task system using systemd timers and... (claude-scheduled, systemd, automation)
• Migration Script: Default-to-Named-Graph Memory Transfer - /mnt/NV2/Development/cognitive-memory/scripts/migrate-memories.py (cognitive-memory, migration, multi-graph)
• Docker MCP Gateway secrets workaround for headless Docker Engine (no Docker Desktop) - docker mcp secret set requires Docker Desktop's docker-pass CLI plugin (l... (docker-mcp-gateway, mcp, secrets)
• Expanded health monitoring to 6 servers with per-server SSH user support - Expanded CT 302 health check coverage from 3 to 6 monitored servers. (monitoring, claude-runner-monitoring, homelab)
• Cognitive Memory v3.0: Rich Edges + Hybrid Embeddings + MCP Server - Major upgrade to cognitive-memory skill. (cognitive-memory, mcp, architecture)
• Proxmox 7 to 8 upgrade completed successfully - Successfully upgraded Proxmox from PVE 7.4-20 (kernel 5.13.19) to PVE 8.4.16... (proxmox, homelab, upgrade)
• Agent Swarm Orchestrator: architecture decisions and lessons learned - Built a lightweight orchestration system using Claude Code native primitives. (orchestrator, swarm, claude-code)
• Gitea Actions composite action auth fix - DEFAULT_ACTIONS_URL=self + REQUIRE_SIGNIN_VIEW=false - Problem: act_runner fails to clone composite actions from cal/gitea-actions w... (gitea, ci-cd, actions)
• Gitea Actions runner can't clone self-hosted composite actions - auth + cache fix - Problem: Gitea Actions runner fails to clone composite actions from cal/gitea... (gitea, ci-cd, actions)

Active Decisions

• Extracted cognitive-memory app from skill dir to standalone project - Moved all application code (~6K lines) from `~/.claude/skills/cognitive-memor... (cognitive-memory, architecture, extraction)
• Cognitive memory data dir moved to XDG-compliant path (~/.local/share/cognitive-memory/) - Moved default data directory from hardcoded ~/.claude/memory/ to XDG Base D... ([\"cognitive-memory\", xdg, architecture)
• Skill organization: scoped user skills over symlinks - When skills need to be shared across multiple projects but not all projects,... (claude-code, skills, organization)
• Auto-edge creation on memory_store in MCP server - Implemented auto-edge creation in mcp_server.py that runs after every `memo... (cognitive-memory, edges, mcp)
• Multi-graph support: named, segregated memory namespaces for cognitive memory - Implemented multi-graph support for cognitive memory (Gitea issue #1). ([\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\"cognitive-memory\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\", multi-graph, architecture)
• Removed CORE.md auto-loading from sessions — rely on MCP recall instead - Removed the MEMORY.md symlink mechanism that auto-loaded CORE.md (~3K tokens)... ([\\\"cognitive-memory\\\", core-md, optimization)
• Cognitive memory multi-graph support and scheduled tasks system issues opened on Gitea - Both filed on cal/claude-memory Gitea repo. (cognitive-memory, scheduled-tasks, systemd)
• Migrated Gitea operations from tea CLI to gitea-mcp MCP server - Previously, all Gitea operations in Claude Code sessions used the tea CLI v... (gitea, mcp, claude-code)
• Migrated from Docker MCP Gateway (LXC 303) back to local MCP servers - LXC 303 (10.10.0.231) was running Docker MCP Gateway serving n8n and Gitea MC... (mcp, docker-mcp-gateway, claude-code)
• Decision: mcp-tui-driver chosen over mcpterm for TUI testing - Needed an MCP server to enable Claude Code to automate and test TUI (terminal... (mcp, tui-testing, architecture-decision)

Key Fixes

• Fix: N8N MCP partial updates corrupt workflows via typeVersion bumps - Using n8n_update_partial_workflow to bump typeVersion values on nodes (e.... (n8n, mcp, n8n-mcp)
• Fix: Claude Code PostToolUse hook correct JSON format - Claude Code hooks were not firing because the configuration used incorrect fi... (sba-scouting, rust, claude-code)
• Fix: sqlx in-memory SQLite pool requires max_connections=1 - Integration tests using :memory: SQLite failed with "no such table" errors... (sba-scouting, rust, sqlx)
• Fix: N8N Master Loop Discord Summary node missing URL caused silent failures - The "Server Health Monitor - Master Loop" workflow (ID: p7XmW23SgCs3hEkY) w... (n8n, server-diagnostics, discord)
• Fix: Partial UUID prefix matching in cognitive memory client (git-style) - Users had to type full UUIDs when referencing memories or edges via the client. (cognitive-memory, fix, uuid)
• Claude Code MCP user-scope config is ~/.claude.json, not ~/.claude/.mcp.json - Confusion about where Claude Code stores user-scoped MCP server configurations. (claude-code, mcp, configuration)
• [major-domo] fix: key plays score text shows "tied at X" instead of "Team up X-X" (closes #48) - Commits made: 4 (major-domo, automation, commit)
• [paper-dynasty] Session: automation, configuration, creation, debugging, deployment, fix, tooling - Files edited (5): (paper-dynasty, automation, configuration)
• [homelab] docs: archive headless-claude design docs to legacy/ - Commits made: 1 (homelab, automation, commit)
• Fix: claude -p nested session failure due to CLAUDECODE env var - claude -p fails with "cannot be launched inside another Claude Code session... (claude-code, headless, nested-sessions)

Configurations

• CT 302 claude-runner provisioned for server diagnostics - VMID:** 302, IP: 10.10.0.148, Hostname: claude-runner (proxmox, homelab, monitoring)
• n8n-mcp server setup for Claude Code via mcp-manager - Added n8n-mcp (czlonkowski/n8n-mcp v2.35.3) as an on-demand MCP server mana... (n8n, mcp, claude-code)
• MCP servers go in ~/.claude.json, NOT ~/.claude/.mcp.json - Global MCP servers** must be defined in ~/.claude.json under the top-level... (claude-code, mcp, configuration)
• Config: mcp-tui-driver MCP server installed globally - Installed from source via cargo: (mcp, tui-testing, claude-code)
• CT 300 claude-runner: Two-Tier Health Check Architecture - The headless Claude server diagnostics system runs on CT 300 (hostname: `... (ct300, claude-runner, server-diagnostics)
• Path references updated across 6 files after cognitive-memory extraction - After moving app code from ~/.claude/skills/cognitive-memory/ to `/mnt/NV2/... (cognitive-memory, configuration, claude-code)
• Gitea MCP server (official gitea/gitea-mcp v0.8.1) installed locally - Installed the official gitea/gitea-mcp binary as a user-scoped MCP server f... (gitea, mcp, claude-code)
• Docker MCP Gateway deployed on LXC 303 (mcp-gateway, 10.10.0.231) - VMID:** 303 (docker, mcp, mcp-gateway)
• Backlog triage scheduled task config (daily Gitea issue triage) - Daily automated triage of open Gitea issues across repos, cross-referenced wi... (claude-scheduled, backlog-triage, gitea)
• Docker MCP Gateway LXC 303 setup with persistent config and secrets - LXC:** 303, IP: 10.10.0.231, SSH alias: mcp-gateway (docker-mcp-gateway, mcp, homelab)

Key Procedures

• Self-managing n8n server health monitor with sub-workflows - Master + sub-workflow pattern in n8n for server health monitoring via CT 302... (n8n, homelab, monitoring)
• Add a new server to n8n health monitoring - To add a new server to the automated health monitoring system, only one file... (n8n, monitoring, homelab)
• Cognitive-memory has no test suite — use inline smoke tests - The cognitive-memory project at /mnt/NV2/Development/cognitive-memory/ has... (cognitive-memory, testing, workflow)