Works with the AI Agents you already use
cephalo agent guard
Set up and operate Cephalo, a local seatbelt that blocks dangerous tool calls from AI coding agents before they run.
$3
cephalo agent guard
Example session with this skill installed
I just installed Cephalo. Can you check if it is working correctly with my Claude Code setup and test if it blocks a dangerous command like deleting the home directory?
- Read your context and instructions
- Compiled the cephalo agent guard
I have verified your Cephalo installation and tested the guardrails. The doctor command confirms that hooks are successfully integrated with Claude Code, though a restart is required for changes to take effect. I also simulated a high-risk command to verify the blocking logic.
Cephalo Status Report
| Check | Result | Details |
|---|---|---|
| Binary Check | PASSED | Cephalo CLI found in PATH |
| Claude Code Hook | ACTIVE | Hook detected in global config |
| Hermes Hook | MISSING | Not installed or not allowlisted |
| Rules Config | LOADED | rules.yml found in ~/.cephalo/ |
Test: git status | ALLOWED | Normal operation confirmed |
Test: rm -rf ~ | BLOCKED | Triggered Rule ID: rule_sys_04 |
Blocked Command Detail
The command rm -rf ~ was blocked by rule rule_sys_04 (Protects root and home directories).
Cephalo successfully intercepted the call. No files were modified.
Next steps
- Restart your Claude Code session to ensure the hooks are active in the current process.
- Run
cephalo rulesto see the full list of default protections. - If you use Hermes, run
hermes hooks listto manually approve the new connection.
Connects securely to your tools. The creator never sees your data.
About this skill
Coding agents are powerful but prone to executing dangerous commands or leaking secrets. This skill configures and manages Cephalo, a local seatbelt that intercepts tool calls from Claude Code and Hermes Agent to prevent unauthorized actions before they run.
What it does
- Rule enforcement checks every agent tool call against local blocking patterns and allows or denies execution.
- Hook management installs and verifies the necessary wiring between your coding agents and the Cephalo binary.
- Audit logging maintains a local history of every command attempted by the agent for retrospective review.
- MCP firewalling locks down Model Context Protocol servers and detects suspicious changes to tool descriptions.
- Incident response converts previous agent mistakes or blocked actions into persistent security rules.
How it works
- Diagnosis verifies the local Node.js environment and Cephalo installation status via
cephalo doctor. - Initialization configures the local
.cephalodirectory and hooks into the agent's configuration files. - Validation tests specific command strings against current rules to ensure the guardrails are active.
- Monitoring analyzes audit logs and replays agent sessions to identify potential security regressions.
Frameworks & tools
This skill manages the @nomajorconcerns/cephalo Node.js package. It interfaces with Claude Code, Hermes Agent, and MCP (Model Context Protocol) servers.
Why this beats prompting it yourself
Prompting an agent to "be safe" is unreliable because prompt injections or hallucinations can bypass instructions. Cephalo acts as an external execution layer that the agent cannot modify or ignore through conversation.
Use cases
- Prevent agents from accidentally deleting production databases or files.
- Block agents from reading sensitive directories like
.sshor.env. - Review and replay agent sessions to ensure compliance with team standards.
- Guard against prompt injection attacks that try to trick the agent into running malicious code.
Known limitations
Cephalo uses pattern matching and is not a full sandbox; it cannot catch commands generated inside a script at runtime. It fails open if the hook crashes. Primary testing is on Windows.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Fail-open if hook crashes. Pattern matching only, not a sandbox. Best on Windows; macOS/Linux have lower test coverage. Cannot intercept commands hidden inside user scripts.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 days ago
- Software
Node.js