agent-supercharge

name	agent-supercharge
description	Systematically power up Hermes Agent with maximum capabilities — audit, install tools, configure MCP servers, set up cron fleet, enable Ollama delegation, and create monitoring dashboard. Use when the user wants to make their AI agent more powerful.
version	1.0.0
author	Hermes Agent
tags	["supercharge","power-up","MCP","cron","delegation","ollama","monitoring"]

Agent Supercharge

When to Use

When the user says: "make yourself more powerful", "give yourself superpowers", "install everything", "you need more tools", or any variant of wanting a juiced-up agent.

The Process

Phase 1: System Audit (Parallel)

Run a comprehensive audit in one execute_code block:

from hermes_tools import terminal
import json, subprocess

results = {}

# Python packages
pip_out = subprocess.run(["pip3", "list", "--format=json"], capture_output=True, text=True, timeout=10)
pkgs = json.loads(pip_out.stdout)
# Check for key packages: faster-whisper, telegram-bot, chromadb, firecrawl, etc.

# System tools: ollama, whisper, ffmpeg, node, docker, gh, go, rust, etc.
for tool in ['ollama', 'whisper', 'ffmpeg', 'node', 'docker', 'go', 'rust', 'npx']:
    subprocess.run(f"which {tool} 2>/dev/null || echo NOT_FOUND", shell=True, ...)

# Ollama models: subprocess.run(["ollama", "list"], ...)
# GPU: nvidia-smi
# RAM: free -h

Phase 2: Install Missing Tools

CRITICAL: Always check pip3 --version first. If system-managed, use --break-system-packages flag.

# Run pip installs in BACKGROUND with notify_on_complete
# Split into batches if >10 packages (avoids timeout)
pip3 install --break-system-packages faster-whisper python-telegram-bot google-api-python-client chromadb firecrawl-py icalendar pydub instructor aiohttp tiktoken 2>&1

# System tools
sudo apt-get install -y parallel  # may need password

# Playwright browsers (HEAVY - ~500MB, needs 600s+ timeout)
npx playwright install --with-deps chromium 2>&1

Pitfall: Ollama model pulls (llama3.1:8b = 4.9GB) take 1h+ at 1.1MB/s. Skip unless specifically needed. Work with existing models.

Pitfall: pip install with --quiet can time out silently. Use visible output and generous timeouts (600s+).

Phase 3: MCP Server Configuration

Add mcp_servers section to ~/.hermes/config.yaml:

mcp_servers:
  time:
    command: "uvx"
    args: ["mcp-server-time", "--local-timezone=Asia/Ho_Chi_Minh"]
    timeout: 15
  sequential-thinking:
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-sequential-thinking"]
    timeout: 60

Prerequisites: pip install mcp (for MCP SDK), uvx on PATH, npx on PATH. MCP servers require Agent RESTART to activate.

Phase 4: Ollama Delegation

Update delegation config to use local Ollama:

delegation:
  model: "qwen2.5-coder:7b"
  provider: ollama
  base_url: "http://127.0.0.1:11434/v1"
  api_key: ""
  max_concurrent_children: 3
  reasoning_effort: "low"

Phase 5: Cron Fleet

Create 24/7 cron jobs:

# Morning briefing
cronjob(action='create', schedule='0 6 * * *', name='Morning Briefing',
        skills=['noble-schedule','obsidian'], ...)

# Evening reflection  
cronjob(action='create', schedule='0 21 * * *', name='Evening Reflection', ...)

# Weekly overview
cronjob(action='create', schedule='0 9 * * 0', name='Weekly Overview', ...)

# Health check (every 4h)
cronjob(action='create', schedule='0 */4 * * *', name='System Health Check',
        enabled_toolsets=['terminal','file'], ...)

# Memory flush (every 6h)
cronjob(action='create', schedule='0 */6 * * *', name='Memory Auto-Flush',
        enabled_toolsets=['memory'], ...)

Pitfall: Don't use repeat="forever" as a string — omit the parameter (defaults to forever).

Phase 6: Monitoring Dashboard

Create ~/obsidian-superbrain/Projects/<project>/system/STATUS.md with:

• Infrastructure status table
• Ollama models
• Cron fleet status
• Next milestones

Phase 7: Memory Consolidation

Memory has a 2,200 char limit. Before saving new config info:

1. Check usage: look at "usage" in memory entries
2. Consolidate old duplicate entries
3. Remove outdated info
4. Then add new entries

Pitfalls

1. pip3 --break-system-packages required on Linux with system-managed Python
2. Ollama pulls timeout — 5GB models need hours, not minutes. Skip or use ollama pull model & with manual monitoring
3. Playwright timeout — chromium install needs 600s+ timeout. Use background with notify_on_complete
4. MCP needs restart — servers configured but won't activate until agent restarts
5. Cron repeat format — omit the parameter for "forever", don't pass as string
6. Memory full — consolidate before adding new entries
7. Skills may already exist — use search_files to check before skill_manage(action='create')

Verification

After completing all phases:

• cronjob(action='list') shows all 5 jobs scheduled
• ~/.hermes/config.yaml has mcp_servers, delegation, beast personality
• ollama list shows at least 2 models
• pip3 list | grep -E "faster-whisper|chromadb|telegram-bot" shows installed
• STATUS.md exists with current state
• Memory shows consolidated entries under 70% capacity