Agentica ServerSAFE

Context management for Claude Code. Hooks maintain state via ledgers and handoffs. MCP execution without context pollution. Agent orchestration with isolated context windows.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: agentica-server
description: Agentica server + Claude proxy setup - architecture, startup sequence, debugging
allowed-tools: [Bash, Read]
user-invocable: false
---

# Agentica Server + Claude Proxy Setup

Complete reference for running Agentica SDK with a local Claude proxy. This enables Python agents to use Claude CLI as their inference backend.

## When to Use

Use this skill when:
- Starting Agentica development with Claude proxy
- Debugging connection issues between SDK, server, and proxy
- Setting up a fresh Agentica environment
- Troubleshooting agent tool access or hallucination issues

## Architecture

```
Agentica SDK (client code)
    | S_M_BASE_URL=http://localhost:2345
    v
ClientSessionManager
    |
    v
Agentica Server (agentica-server)
    | INFERENCE_ENDPOINT_URL=http://localhost:8080/v1/chat/completions
    v
Claude Proxy (claude_proxy.py)
    |
    v
Claude CLI (claude -p)
```

## Environment Variables

| Variable | Set By | Used By | Purpose |
|----------|--------|---------|---------|
| `INFERENCE_ENDPOINT_URL` | Human | agentica-server | Where server sends LLM inference requests |
| `S_M_BASE_URL` | Human | Agentica SDK client | Where SDK connects to session manager |

**KEY:** These are NOT the same endpoint!
- SDK connects to server (port 2345)
- Server connects to proxy (port 8080)

## Startup Sequence

Must start in this order (each in a separate terminal):

### Terminal 1: Claude Proxy

```bash
uv run python scripts/agentica/claude_proxy.py --port 8080
```

### Terminal 2: Agentica Server

**MUST run from its directory:**

```bash
cd workspace/agentica-research/agentica-server
INFERENCE_ENDPOINT_URL=http://localhost:8080/v1/chat/completions uv run agentica-server --port 2345
```

### Terminal 3: Your Agent Script

```bash
S_M_BASE_URL=http://localhost:2345 uv run python scripts/agentica/your_script.py
```

## Health Checks

```bash
# Claude proxy health
curl http://localhost:8080/health

# Agentica server health
curl http://localhost:2345/health
```

## Common Errors & Fixes

### 1. APIConnectionError after agent spawn

**Symptom:** Agent spawns successfully but fails on first call with connection error.

**Cause:** Claude proxy returning plain JSON instead of SSE format.

**Fix:** Proxy must return Server-Sent Events format:
```
data: {"choices": [...]}\n\n
```

### 2. ModuleNotFoundError for agentica-server

**Symptom:** `ModuleNotFoundError: No module named 'agentica_server'`

**Cause:** Running `uv run agentica-server` from wrong directory.

**Fix:** Must `cd workspace/agentica-research/agentica-server` first.

### 3. Agent can't use Read/Write/Edit tools

**Symptom:** Agent asks for file contents instead of reading them.

**Cause:** Missing `--allowedTools` in claude_proxy.py CLI call.

**Fix:** Proxy must pass tool permissions:
```bash
claude -p ... --allowedTools Read Write Edit Bash
```

### 4. Agent claims success but didn't do task

**Symptom:** Agent says "I've created the file" but file doesn't exist.

**Cause:** Hallucination - agent describing intended actions without executing.

**Fix:** Added emphatic anti-hallucination prompt in REPL_BASELINE:
```
CRITICAL: Use ACTUAL tools. Never DESCRIBE using tools.
```

### 5. Timeout on agent.call()

**Symptom:** Call hangs for 30+ seconds then times out.

**Cause:** Claude CLI taking too long or stuck in a loop.

**Fix:** Check proxy logs for the actual CLI output. May need to simplify prompt.

## Key Files

| File | Purpose |
|------|---------|
| `scripts/agentica/claude_proxy.py` | OpenAI-compatible proxy with SSE streaming |
| `workspace/agentica-research/agentica-server/` | Local agentica-server installation |
| `scripts/agentica/PATTERNS.md` | Multi-agent pattern documentation |

## Quick Verification

Test the full stack:

```bash
# 1. Verify proxy responds
curl http://localhost:8080/health

# 2. Verify server responds
curl http://localhost:2345/health

# 3. Test inference through proxy
curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"claude","messages":[{"role":"user","content":"Say hello"}]}'
```

## Checklist

Before running agents:
- [ ] Claude proxy running on port 8080
- [ ] Agentica server running on port 2345 (from its directory)
- [ ] `S_M_BASE_URL` set for client scripts
- [ ] `INFERENCE_ENDPOINT_URL` set for server
- [ ] Both health checks return 200

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

.claude/scripts/mcp

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: Agentica Server](https://modelsatlas.com/skills/parcadei/agentica-server/badge.svg)](https://modelsatlas.com/skills/parcadei/agentica-server/)

Not your listing, or something out of date? Claim or update your listing.