SHIELD ALARMS

Any attempt to authenticate to an LLM provider using the alarm key triggers an alert. Covers direct API calls, agents reading credentials from .env files or context windows, and automated tooling that sweeps for usable keys.

Shield generates a key that matches the format of real provider keys (OpenAI, Anthropic, etc.). Place it in your .env file, agent configuration, or any location where real credentials live. The alarm is registered against Shield's intercept endpoint before it can be used.

Any on-chain interaction with the wallet address triggers an alert, including transfer attempts, contract calls, or balance checks from non-whitelisted addresses. Useful for detecting agents or attackers attempting to exfiltrate funds.

Shield provides a monitored wallet address on your chosen chain (Ethereum, Solana, or Bitcoin). Include the address in agent tool configurations, smart contract whitelists, or anywhere a legitimate wallet address might appear in your infrastructure.

Any connection attempt using the honeypot credentials triggers an alert, including automated credential stuffing, lateral movement by a compromised agent, or an attacker who has read your database configuration.

Shield provides a set of fake database credentials (host, port, user, password, database name). Add them to your .env file, secrets manager, or any configuration file alongside real credentials. Shield listens for connection attempts on the specified endpoint.

File read events on the monitored path trigger an alert. This catches agents or processes that scan the filesystem for credential files, as well as attackers who have gained code execution and are searching for secrets.

Shield's agent installs a filesystem watcher on the path you specify. Place a realistic-looking .env file at that path. Any process that opens it for reading generates an alert with the process name, PID, and user.

Any TCP or UDP connection attempt to the monitored port triggers an alert, enriched with GeoIP data, ASN, and reverse DNS. Useful for detecting network probes, lateral movement, and unauthorized port scanning.

Choose a port that should never receive legitimate traffic in your environment. Shield's agent opens a listener on that port. You can also deploy listeners on well-known ports like 22 (SSH), 3306 (MySQL), or 5432 (PostgreSQL) by using the agent's port mirroring mode, which does not interfere with existing services.

Any SSH authentication attempt using the honeypot key triggers an alert with the source IP, port, username, and key fingerprint. Catches attackers who have found a private key and are attempting to use it for lateral movement.

Shield generates a private key and registers the corresponding public key against a monitored endpoint. Place the private key in ~/.ssh/, agent configurations, or any location where SSH keys are typically stored. The key is cryptographically valid and will authenticate, triggering the alert on first use.

Any agent invocation of the honeypot tool triggers an alert, including the full tool call context: arguments, calling agent identity, and session information. This is particularly effective for detecting prompt injection attacks where a malicious instruction directs an agent to call a specific tool.

Shield registers a honeypot MCP tool under a name of your choosing. Add it to your MCP server configuration so agents can see it. Give it a plausible name that a malicious prompt might target (for example, a tool that appears to exfiltrate data or bypass a restriction). Any agent that invokes it triggers an alert.