A multi-agent deliberation hub with 3 core agents and extensible extended agents.
hello, hi, hey, thanks, thank you
what time, weather, temperature
yes, no, ok, sure
define, what is
Query → Check Skip List
│
▼ (if not skipped)
┌──────────────────┐
│ CORE AGENTS │ (always run)
│ - Intent Decoder│
│ - Risk Checker │
│ - Tone Designer │
└──────────────────┘
│
▼ (if extended triggered)
┌──────────────────┐
│ EXTENDED AGENTS │ (included)
└──────────────────┘
│
▼
┌──────────────────┐
│ WORKSPACE SKILLS │ (if needed)
└──────────────────┘
│
▼
Enriched Response
What does the user actually want?
What could go wrong?
How should this feel?
| Agent | Trigger Keywords |
|---|---|
| ------- | ----------------- |
| System Designer | api, database, architecture, system |
| Complexity Assessor | complex, analyze, compare |
| Values Guardian | ethical, moral, values, fair |
The council can call skills from your workspace when specialized knowledge is needed.
Example:
To add workspace skills:
workspace/skills/workspace/agents/Create a new .md file in agents/ folder:
# Your Agent Name
Trigger: keyword1, keyword2
Your analysis...
Skill runs silently. Apply analysis to shape response.
Use council: prefix to see full analysis.
IF query contains: dangerous, risk, worried, scared
THEN: Risk Checker flag = high
IF query contains: frustrated, angry, upset
THEN: Tone = empathetic
Keywords: multi-agent, AI router, agent council, decision support, AI deliberation, extensible hub, workspace skills
Use cases: personal AI assistant, decision making, risk assessment
共 1 个版本