Works with the AI Agents you already use
SteerPlane Agent Guardrails
This skill teaches your coding agent (Claude Code, Cursor, Codex and others) to add SteerPlane to the agent you are building, so every run has hard limits.
Free
SteerPlane Agent Guardrails
Example session with this skill installed
I have a LangChain agent using gpt-4o-mini. I need to make sure it never spends more than $2.00 per run and stays under 30 steps. Add SteerPlane to protect it.
- Read your context and instructions
- Compiled the steerplane agent guardrails
I have integrated SteerPlane into your LangChain agent. I added the SteerPlaneCallbackHandler to your LLM configuration and defined the requested cost and step limits. I also included a basic exception handler to catch when these limits are hit.
SteerPlane Integration for LangChain
from langchain_openai import ChatOpenAI
from steerplane.integrations.langchain import SteerPlaneCallbackHandler
from steerplane.exceptions import CostLimitExceeded, StepLimitExceeded
# 1. Initialize the SteerPlane handler with your constraints
sp_handler = SteerPlaneCallbackHandler(
agent_name="langchain_protected_bot",
max_cost_usd=2.0,
max_steps=30,
)
# 2. Attach the handler to the LLM
llm = ChatOpenAI(
model="gpt-4o-mini",
callbacks=[sp_handler]
)
# 3. Example of handling the enforcement at runtime
def run_protected_agent(prompt):
try:
# Assuming 'agent_executor' is your initialized LangChain agent
return agent_executor.invoke({"input": prompt})
except CostLimitExceeded:
return "Agent stopped: Per-run budget of $2.00 reached."
except StepLimitExceeded:
return "Agent stopped: Maximum of 30 steps reached."
Next steps
- Run
pip install steerplanein your terminal. - Update your agent initialization code to pass the
llminstance with the new callbacks. - Test a long-running prompt to verify the
StepLimitExceededexception triggers correctly.
Connects securely to your tools. The creator never sees your data.
About this skill
What this skill does
Your agent can loop on the same tool, keep retrying, or get left running. You usually find out from the bill. This skill teaches your coding agent (Claude Code, Cursor, Codex and others) to add SteerPlane to the agent you are building, so every run has hard limits.
What it sets up
- Cost limit per run (
max_cost_usd). The run stops when it is reached. - Step and time limits (
max_steps,max_runtime_sec). - Loop detection for agents that repeat the same actions.
- Allow / deny rules with glob patterns, for example
denied_actions=["delete_*", "sudo_*"]. - Clean error handling, so a stopped run shows a clear message instead of a crash.
Works with
- Python agents (decorator or context manager)
- LangChain (callback handler)
- OpenAI Agents SDK (agent hooks)
- TypeScript agents
Why it's free
SteerPlane is open source (MIT). pip install steerplane or npm install steerplane. Everything runs locally. The skill only guides your agent through the setup and never sends your code anywhere.
Docs: https://docs.steerplane.com Code: https://github.com/vijaym2k6/SteerPlane
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Loop detection may flag legitimate repetitive actions. Requires manual integration into the agent execution loop.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 17 days ago
- Free with an Agensi account
- Software
Python
Permissions
No special permissions declared or detected