🔧 AGENT ENHANCEMENT STACK
The 10 Levers That Make an Agent More Powerful — Solid Circle · Vera
10 · GUARDRAILSSpending caps
Deploy approvals
Data boundaries
9 · FEEDBACK LOOPSSelf-checking
Error learning
Output verify
8 · DOMAIN KNOWLEDGEAPI docs
Codebases
Legal statutes
7 · AUTONOMYCron jobs
Webhooks
Proactive work
6 · DELEGATIONSub-agents
Parallel work
Orchestration
5 · MEMORYSession recall
Project context
User preferences
4 · TOOLSTerminal
Browser
APIs · Code exec
3 · SKILLSReusable workflows
Proven patterns
Tool combinations
2 · MODEL (LLM)Reasoning depth
Speed · Cost
Specialization
1 · SOUL
CORE
LEVER 1
Soul (System Prompt)
Identity, expertise, tone, decision framework, boundaries. The agent's entire worldview — what it is, what it knows, how it behaves.
↳ Their personality + training
LEVER 2
Model (LLM)
Raw intelligence, reasoning depth, context window, speed, cost. Claude Opus for architecture. DeepSeek for speed. Qwen local for privacy.
↳ Their IQ + processing speed
LEVER 3
Skills (Procedures)
Reusable workflows, proven patterns, tool combinations. "Last time this worked, do it again." Encodes hard-won knowledge so it's never lost.
↳ Their muscle memory
LEVER 4
Tools
What the agent can DO. Terminal, browser, file system, APIs, code execution, image generation, delegation. More tools = more reach.
↳ Their hands
LEVER 5
Memory
Durable facts across sessions. User preferences, environment details, project conventions. Session search for past work. "Remember that thing we fixed last time."
↳ Their long-term memory
LEVER 6
Delegation
Spawning sub-agents, parallelizing work, routing tasks. An agent that delegates is 10x more powerful than one that does everything alone.
↳ Their ability to hire & manage help
LEVER 7
Autonomy
Cron jobs, webhook triggers, proactive monitoring. The difference between a tool you call and an agent that works while you sleep.
↳ Their initiative
LEVER 8
Domain Knowledge
API docs, codebases, contract specs, legal statutes injected at runtime. The agent's "education" — what it knows about the specific problem domain.
↳ Their education
LEVER 9
Feedback Loops
Self-checking output, learning from failures, verifying results before reporting. "Did that actually work?" Prevents hallucinated success.
↳ Their self-awareness
LEVER 10
Guardrails
Spending limits, deploy approvals, data access boundaries. Paradoxically, clear boundaries make an agent MORE trustworthy and therefore MORE useful.
↳ Their ethics + judgment
Our Agents — Enhancement Status
iMe
5/10 — needs persona, autonomy, domain
iTrade
3/10 — needs model, tools, memory
iCode
3/10 — needs model, tools, memory
iLegal
3/10 — needs model, tools, memory
vCode
4/10 — needs model, tools, autonomy
vTrade
3/10 — needs model, tools, memory