Create or update a security policy for agent governance.

Policies define rules that control what agents can do. Each policy contains conditions and actions (allow/deny/require_approval).

Args: policy_id: Unique identifier for the policy (e.g., "prod-db-access") name: Human-readable name (e.g., "Production Database Access Control") description: Description of what this policy controls rules: JSON array of rule objects. Each rule has: - condition: Object with matching criteria - tool_pattern: Glob pattern for tool names (e.g., "database_*") - action_type: Type of action (e.g., "tool_call") - trust_level_at_least: Minimum trust level required - trust_level_below: Trigger if trust below this level - action: "allow", "deny", or "require_approval" - message: Message to show when rule matches priority: Higher priority policies are evaluated first (default: 100) enabled: Whether the policy is active (default: true)

Returns: JSON string with creation result: - success: Whether creation succeeded - policy_id: The policy's ID - message: Status message

Example: create_policy( policy_id="prevent-deletions", name="Prevent Dangerous Deletions", description="Block delete operations for non-admin agents", rules='[{"condition": {"tool_pattern": "delete_*", "trust_level_below": "admin"}, "action": "deny", "message": "Delete operations require admin access"}]' )

Retrieve audit log entries for compliance and investigation.

The audit log records all action validations, policy violations, and administrative actions performed through Guardian Agent.

Args: agent_id: Filter by specific agent ID (optional) action_type: Filter by action type like "tool_call", "resource_access" (optional) time_range: Time range to query - "1h", "24h", "7d", "30d" (default: "24h") status: Filter by status - "allowed", "denied", or "" for all (optional) limit: Maximum number of entries to return (default: 100)

Returns: JSON string with: - entries: Array of audit log entries - count: Number of entries returned - total: Total entries matching filter (before limit) - time_range: The time range used - filters_applied: Summary of filters used

Example: # Get all denied actions in the last hour get_audit_log(time_range="1h", status="denied")

# Get all actions by a specific agent
get_audit_log(agent_id="prod-agent-01", time_range="7d")

Get compliance status and security health metrics.

Generates a compliance report showing policy violations, security incidents, and overall governance health for the specified time period.

Args: time_range: Time range - "1h", "24h", "7d", "30d" (default: "24h") include_incidents: Include active incidents in report (default: true) include_policy_summary: Include policy overview (default: true)

Returns: JSON string with compliance report: - status: Overall status ("healthy", "warning", "critical") - metrics: Key security metrics - incidents: Active security incidents (if requested) - policies: Policy summary (if requested) - recommendations: Suggested actions to improve security

Example: # Get daily compliance status get_compliance_status(time_range="24h")

# Get weekly report with all details
get_compliance_status(time_range="7d", include_incidents=True)

Register a new agent with Guardian for security policy evaluation.

Registered agents get proper trust levels and can have custom tool permissions. Unregistered agents are treated as 'low' trust.

Args: agent_id: Unique identifier for the agent (e.g., "prod-assistant-01") name: Human-readable name (e.g., "Production Assistant") description: Description of the agent's purpose trust_level: Trust level - "low", "medium", "high", or "admin" allowed_tools: JSON array of allowed tool patterns (e.g., '["read_*", "query_*"]') denied_tools: JSON array of denied tool patterns (e.g., '["delete_*", "drop_*"]') metadata: JSON object with additional agent metadata

Returns: JSON string with registration result: - success: Whether registration succeeded - agent_id: The agent's ID - message: Status message - warnings: Any warnings about the registration

Example: register_agent( agent_id="data-analyst-01", name="Data Analyst Bot", description="Runs analytical queries on warehouse", trust_level="medium", allowed_tools='["query_*", "read_*"]', denied_tools='["delete_*", "drop_*", "truncate_*"]' )

Report a security incident for investigation and tracking.

Use this tool to log security incidents such as policy violations, suspicious agent behavior, or potential security threats.

Args: incident_type: Type of incident: - "policy_violation": Agent violated a security policy - "suspicious_activity": Unusual or potentially malicious behavior - "unauthorized_access": Attempt to access restricted resources - "rate_limit_exceeded": Agent exceeded rate limits - "data_exfiltration": Potential data leak detected - "configuration_error": Security misconfiguration detected - "other": Other security concern severity: Incident severity: - "low": Minor issue, no immediate action needed - "medium": Notable issue, should be reviewed - "high": Serious issue, needs prompt attention - "critical": Emergency, immediate action required description: Detailed description of the incident agent_id: ID of the agent involved (if applicable) evidence: JSON object with supporting evidence/data recommended_action: Suggested remediation steps

Returns: JSON string with: - incident_id: Unique incident identifier - success: Whether the incident was logged - message: Status message - agent_suspended: Whether the agent was auto-suspended

Example: report_incident( incident_type="suspicious_activity", severity="high", description="Agent attempted to access 50 databases in 1 minute", agent_id="rogue-agent-01", evidence='{"databases_accessed": 50, "time_window": "60s"}', recommended_action="Review agent permissions and suspend if needed" )

Validate whether an agent can perform a specific action.

This is the PRIMARY security gate. Agents should call this BEFORE performing any sensitive action to ensure compliance with security policies.

Args: action_type: Type of action (e.g., "tool_call", "resource_access", "data_read", "data_write") target: Target of the action (e.g., tool name, resource URI, database name) agent_id: Unique identifier of the requesting agent parameters: JSON string of action-specific parameters (optional) context: Additional context about why this action is needed (optional)

Returns: JSON string with validation result: - action_id: Unique ID for this validation (for audit correlation) - allowed: Whether the action is permitted - require_approval: If true, action needs human approval first - reason: Explanation of the decision - warnings: Any non-blocking warnings

Example: validate_action( action_type="tool_call", target="database_delete", agent_id="prod-agent-01", parameters='{"table": "users"}', context="Cleanup stale records" )