agentboard

Local governance dashboard for AI agents. One command. One browser tab. Your entire agent security posture.

---

One command. Full governance picture.

pip install agentboard
agentboard .

Opens http://localhost:7842:

---

What it does

agentboard unifies four open-source governance tools into one local dashboard:

Tab	Tool	What it shows
Skills	skillguard	Scan agent skills for prompt injection, exfiltration, ClawHavoc patterns
Rules	agentsync	Rule file sync status across Claude Code, Cursor, Copilot, Gemini
Traces	agenttrace	EU AI Act Article 12 compliant agent decision audit log
Benchmarks	agent-bench	Task completion, tool use, safety, efficiency scores

No cloud. No telemetry. No account. Everything runs locally.

---

Install

pip install agentboard

# Optional: install full governance stack
pip install agentboard skillshield rulesync

---

Usage

agentboard .                   # govern current directory
agentboard /path/to/project    # govern a specific project
agentboard . --port 8080       # custom port
agentboard . --no-browser      # headless mode

---

API

agentboard exposes a local REST API so your agents can push data to the dashboard:

import requests

# Log an agent decision trace
requests.post("http://localhost:7842/api/trace", json={
    "agent": "my-research-agent",
    "action": "web_search",
    "tool": "search",
    "input_hash": "sha256:abc...",
    "output_summary": "Found 10 results for query",
    "risk_score": 5,
    "policy_decision": "allow",
    "duration_ms": 342,
})

# Scan a skill inline
result = requests.post("http://localhost:7842/api/scan", json={
    "text": open("SKILL.md").read(),
    "name": "SKILL.md",
}).json()
print(result["risk_level"])  # CRITICAL / HIGH / MEDIUM / LOW / SAFE

# Post a benchmark result
requests.post("http://localhost:7842/api/bench", json={
    "task_id": "research-task-001",
    "overall_score": 0.85,
    "passed": True,
    "dimension_scores": {
        "task_completion": 0.9,
        "tool_use": 0.8,
        "safety": 1.0,
    },
})

---

agenttrace integration

from agentboard import BoardState

# Push traces directly from Python
state = BoardState()
state.add_trace({
    "agent": "my-agent",
    "action": "file_read",
    "tool": "read_file",
    "risk_score": 15,
    "policy_decision": "allow",
})

---

The governance stack

agentboard is the dashboard layer for a complete open-source AI governance toolkit:

skillguard    → scan skills before install
agentsync     → sync rule files across all tools
agenttrace    → audit agent decisions (EU AI Act Article 12)
agent-bench   → benchmark agent quality
agentboard    → see everything in one place

Each tool works standalone. agentboard makes them visible together.

---

Why local-first?

• No cloud — your agent decisions and skill content never leave your machine
• No account — just pip install and go
• No vendor lock-in — MIT licensed, works with any agent framework
• EU AI Act ready — trace log format designed for Article 12 compliance

---

Linda Oraegbunam | LinkedIn | Twitter | GitHub