🤖 Agent Society Integration

Use LibreChat and Zed IDE to orchestrate multi-agent workflows.

Overview

The Agent Society provides governed AI agent orchestration:

Component Via LibreChat Via Zed IDE
Configure agents Chat prompts File editing + validation
Run workflows Conversational Command palette
Query results Natural language SPARQL/structured
Audit review Ask for logs Agent panel output

Agent Configuration via Chat

Create an Agent

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

You: Create a code review agent with these settings:
- Role: R-AA (Automated Agent)
- Sponsor: R-DS:jane.doe
- Skills: review-code, flag-issues
- Cannot approve own recommendations

SEA™: I'll configure the agent. Here's the spec:

```yaml
agentId: "code-reviewer"
name: "Code Review Agent"
role: "R-AA"
sponsor: "R-DS:jane.doe"
capabilities:
  - review-code
  - flag-issues
restrictions:
  - cannot_approve_own_recommendations
  - requires_hitl_for_mutations

1
2
3
4
5
6
7
8
9
10
11
12
13

### Deploy via Zed

1. Save agent config to `agents/code-reviewer.agent.yaml`
2. Cmd+Shift+V to validate
3. Run: `just agent-deploy code-reviewer`

---

## Cognitive Workflows

### Router + Specialists Pattern

You: Analyze the quarterly report using:

Use the Softmax Router pattern (SDS-006).

SEA™: Starting cognitive workflow…

Round 1: ├── Router: Delegating to 3 specialists ├── Semantic Mapper: Extracted 12 concepts ├── Rule Analyst: 2 potential violations └── Architecture Checker: All boundaries valid

Round 2: ├── Router: Aggregating results └── Convergence: 97% (threshold: 95%)

Final Output: { “concepts”: […], “violations”: […], “architecture_status”: “compliant” }

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

---

## HITL (Human-in-the-Loop) Escalation

### Trigger Conditions

| Condition | Example | Channel |
|-----------|---------|---------|
| Mutation requests | Code changes | LibreChat approval |
| Authority boundary | Existential tier | Slack/Teams notification |
| Policy violation | SDS-031 breach | Blocks until human review |

### In LibreChat

SEA™: This action requires human approval:

[Approve] [Reject] [Escalate]

1
2
3
4
5
6
7

---

## Governance Queries

### Audit Log via Chat

You: Show me all actions by agent code-reviewer in the last 24 hours

SEA™: Agent Activity Report: | Time | Action | Status | Sponsor | |——|——–|——–|———| | 09:00 | review-code | ✅ | R-DS:jane.doe | | 09:15 | flag-issues | ✅ | R-DS:jane.doe | | 09:30 | propose-change | ⏳ HITL | pending |

1
2
3
4
5
6
7
8
9
10

### Authority Check via SPARQL

```sparql
SELECT ?agent ?role ?sponsor WHERE {
  ?agent a sea:Agent ;
         sea:role ?role ;
         sea:sponsor ?sponsor .
  FILTER(?role = "R-AA")
}
Task Prompt
List agents “Show all configured agents and their roles”
Check authority “Can agent X perform action Y? Cite policy”
View workflows “List active cognitive workflows”
Debug agent “Why did agent X fail on request Y?”

See Also:

Last Updated: January 2026