* Create a new approval request.

Create a complete compliance suite with multiple frameworks. Returns: Dictionary with compliance engine and constitutional AI

Convenience function to create a governed MCP server. Args: control_plane: Agent Control Plane instance agent_id: Agent

Factory function for creating isolated signal dispatchers.

Create a complete ML-based safety suite. Returns: Dictionary with jailbreak detector and anomaly detector

Create a complete multimodal capabilities suite. Returns: Dictionary with vision, audio, and RAG capabilities

Create a complete observability suite. Returns: Dictionary with all observability components

Create security policies with SQL rules driven by external config. Load SQL policy rules from a YAML config file or a pre-built ``SQLPolicyCo

Factory function to create a PolicyReward with custom configuration. Args: kernel: Agent OS KernelSpace base_reward_fn: Base

Create an agent with read-only permissions

Example: Create capabilities for a SQL-generating agent This agent can only query databases, not modify them. If asked to "build a r

Auto-select a diverse, representative subset of golden traces. Strategy ``"diverse"`` picks traces with different label combinations

De-identify text by replacing PHI with placeholders.

Disable this skill's shell interception activation.

Create a default staged rollout spec.

Delete a file.

Delete the keypair for *agent_id*. Args: agent_id: Unique identifier of the agent. Raises: KeyError: If no k

Remove a trust score. Args: agent_did: DID of the agent. Returns: True if the score existed and was removed.

Deny a pending tool call by its call_id.

Return a denial for a memory-write blocked-pattern match.

Return a denial for an output blocked-pattern match.

Return a denial for behavioral drift above the configured threshold.

Mark a tool version as deprecated. Args: name: Tool name version: Version to deprecate message: Optional

Remove an agent from the fleet.

Detect jailbreak attempts in text. Args: text: Text to analyze context: Additional context (previous

* Detect performance issues in code

Return the names of detected credential patterns. Args: value: String content to inspect. Returns: A de-dupl

Detect if data contains PHI.

Detect the structural characteristics of a document. Returns information about: - Section hierarchy - Key se

* Diagnose an error from error message/stack

Disable agent identity (requires admin to re-enable).

Gracefully close the connection.

Discover agents matching criteria. Args: capabilities: Required capability domains min_score: Minimu

Run a full compatibility check on all installed governance packages. Discovers installed packages, checks every pair against the compatibilit

Convert a declarative PolicyDocument back to a GovernancePolicy.

Deliver a batch of governance events to the backend.

Emit an "Abandonment" signal - Loss. This is called when a user starts a workflow but stops responding halfway witho

Emit an "Acceptance" signal - Success. This is called when a user takes the output and moves to the next task withou

Emit all spans to a LightningStore. Args: store: Agent-Lightning LightningStore instance Returns: N

Emit an "Undo" signal - Critical Failure. This is called when a user reverses an agent action (e.g., Ctrl+Z, revert code).

Mark a plugin version as end-of-life (no longer installable).

End the current Agent Shield session.

* Ensure data directory exists.

* Ensure data directory exists.

Return the bare envelope dict for ``intervention_point``. Convenience for callers that want to build a custom intervention point payl

Run *detector* over *dataset* and return aggregated metrics.

Evaluate trust policies against *context* with async + thread safety. Async ``evaluate`` calls are intentionally serialised with an `

Evaluate a cross-organizational interaction. Steps: 1. Same-org shortcut — bypass federation if same org. 2. Check blocklists

Evaluate an action against policy. Returns GovernanceDecision.

Evaluate whether agent delegation should be allowed.

Evaluate and, if escalated, block until resolved. Convenience method that combines ``evaluate()`` and ``resolve()``.