llama-recourse

LlamaIndex tools for RecourseOS - evaluate consequences before your AI agent executes destructive actions.

Installation

pip install llama-recourse

Requires: Node.js 18+ (for npx recourse-cli)

Quick Start

from llama_recourse import get_recourse_tools
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI

# Get RecourseOS tools
tools = get_recourse_tools()

# Create agent with consequence checking
llm = OpenAI(model="gpt-4")
agent = ReActAgent.from_tools(
    tools,
    llm=llm,
    verbose=True,
    system_prompt="""Before any destructive action, use recourse_evaluate_* tools.
    If risk is BLOCK, refuse to proceed. If ESCALATE, ask user to confirm."""
)

response = agent.chat("Delete the S3 bucket prod-backups")

Tools

recourse_evaluate_terraform

Evaluate Terraform plans before terraform apply.

from llama_recourse import recourse_evaluate_terraform

result = recourse_evaluate_terraform(
    plan_json='{"resource_changes": [...]}',
    state_json=None  # optional
)
print(result)
# **Risk Assessment: BLOCK**
# ...

recourse_evaluate_shell

Evaluate shell commands before execution.

from llama_recourse import recourse_evaluate_shell

result = recourse_evaluate_shell("aws s3 rm s3://prod-data --recursive")
print(result)
# **Risk Assessment: BLOCK**
# ...

recourse_evaluate_mcp

Evaluate MCP tool calls before invocation.

from llama_recourse import recourse_evaluate_mcp

result = recourse_evaluate_mcp(
    server="aws",
    tool="s3.delete_bucket",
    arguments={"bucket": "prod-data"}
)
print(result)
# **Risk Assessment: ESCALATE**
# ...

Full Agent Example

from llama_recourse import get_recourse_tools
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI
from llama_index.core.tools import FunctionTool

# Your existing tools
def execute_shell(command: str) -> str:
    """Execute a shell command."""
    import subprocess
    result = subprocess.run(command, shell=True, capture_output=True, text=True)
    return result.stdout or result.stderr

shell_tool = FunctionTool.from_defaults(
    fn=execute_shell,
    name="execute_shell",
    description="Execute a shell command"
)

# Combine with RecourseOS tools
all_tools = [shell_tool] + get_recourse_tools()

# Create safety-aware agent
agent = ReActAgent.from_tools(
    all_tools,
    llm=OpenAI(model="gpt-4"),
    verbose=True,
    system_prompt="""You are a DevOps assistant.

CRITICAL: Before using execute_shell with ANY destructive command,
you MUST first use recourse_evaluate_shell to check consequences.

Based on the risk assessment:
- BLOCK: Refuse to proceed. Explain the danger.
- ESCALATE: Ask for explicit user confirmation.
- WARN: Inform user of risks, proceed if they agree.
- ALLOW: Proceed normally."""
)

# The agent will now check consequences before destructive actions
response = agent.chat("Remove all files from /tmp/old-backups")
print(response)

With Query Engine Tools

from llama_recourse import get_recourse_tools
from llama_index.core.agent import ReActAgent
from llama_index.core.tools import QueryEngineTool

# Your query engine
query_tool = QueryEngineTool.from_defaults(
    query_engine=index.as_query_engine(),
    name="docs_search",
    description="Search documentation"
)

# Add RecourseOS for safety
tools = [query_tool] + get_recourse_tools()
agent = ReActAgent.from_tools(tools, llm=llm)

Risk Levels

Requirements

• Python 3.9+
• Node.js 18+ (for npx recourse-cli)
• llama-index-core>=0.10.0

License

MIT