Core Agent Toolset

The Core Agent is OLAV's "all-around engineer" — it can execute Python code, SQL queries, shell commands, and web searches. When other Agents fall short, the Core Agent has your back.

olav --agent core "your request"

Feature Claims

ID	Claim	Status
C-L2-31	Core Agent executes Python code and returns output	✅ v0.10.0
C-L2-32	Core Agent executes SQL queries against the business database	✅ v0.10.0
C-L2-33	Core Agent performs web searches and summarizes results	✅ v0.10.0
C-L2-34	Core Agent executes shell commands	✅ v0.10.0
C-L2-35	Config Agent detects workspace health issues	✅ v0.10.0

---

Executing Python Code

Have the Agent run Python directly — great for data analysis, calculations, and quick scripts:

olav --agent core "run this python: import sys; print(sys.version)"

🔧 run_python_code(...)
Python 3.12.3 (main, Mar 3 2026, ...)

More examples:

olav --agent core "计算这组数据的 P95 延迟: [12, 45, 23, 89, 34]"
olav --agent core "解析这个 CSV 并统计 A 列的唯一值数量"
olav --agent core "画一个柱状图: red=10, blue=25, green=15 并保存到 /tmp/chart.png"

Python runs within the OLAV platform's Python environment; output is captured and displayed in the response.

---

Executing SQL Queries

Query business data directly in the domain database:

olav --agent core "what tables are in the database?"

Main Tables (use app. prefix):
 • app.assets (name, type, status, owner)
 • app.events (asset_name, event_type, event_data, snapshot_id)
 • app.relationships (source, target, relationship_type, ...)

Views (no prefix needed):
 • v_asset_summary, v_recent_events, v_active_alerts, ...

You can describe what you need in natural language or write SQL directly:

olav --agent core "how many asset records are in the database?"
olav --agent core "SELECT name, type FROM app.assets"
olav --agent core "show the 5 most recent events"

SQL queries target the business database

The Core Agent's execute_sql connects to .olav/databases/domain.duckdb (business data). To query audit logs, use the olav log command or connect directly to .olav/databases/audit.duckdb.

---

Web Search

When you need to look up external documentation or technical references:

olav --agent core "search: Python asyncio best practices"
olav --agent core "search: DuckDB 窗口函数教程"

The Agent searches the web, extracts relevant content, and returns a summary.

---

Shell Commands

Execute system commands to check server status or files:

olav --agent core "run: df -h"
olav --agent core "run: ls -la exports/"
olav --agent core "检查 2280 端口是否开放"

Mind the execution path

Shell commands run as the current user in the OLAV project root directory. Use absolute paths or explicit relative paths.

---

Skipping Confirmation

In automation scenarios, skip the confirmation prompt for each tool invocation:

olav --agent core --auto-approve "run: df -h"

---

Workspace Health Check (Config Agent)

The Config Agent can diagnose common workspace issues:

olav --agent config "执行健康检查"
olav --agent config "审计工作空间是否有错误"

Issues it can detect:

• @tool function name conflicts (multiple skills defining tools with the same name)
• MANIFEST.yaml missing required fields
• AGENT.md referencing non-existent files

---

Core Agent vs Config Agent: Which One to Use?

Task	Which Agent
Execute Python / SQL / Shell	core
Web search	core
Data analysis and calculations	core
Register external services	config
Workspace health check	config
Install / generate skills	config