Create a GovernanceStepCallback from the kernel.

Ingest a raw signal and normalize it to a ContextObject. This is the main entry point. ANY signal type can be ingested here.

Lint a single YAML policy file and return structured results.

Helper to create an ExecutionRequest for a SQL query.

Valid registration/update signature for the given manifest.

Reset environment for new episode. Args: seed: Random seed (for compatibility) options: Additional options

Search documents by content or metadata with optional time-based decay ranking. When time decay is enabled: - Recent

Get aggregate fleet health status.

Check if a task should be allowed to proceed. Returns (allowed, reason). WARNING — this is an advisory check. ``check_task`` does NO

Decrypt a received message. Handles DH ratchet steps and out-of-order message delivery via cached skipped message keys. Args

Evaluate a task envelope against trust policies. Checks (in order): 1. Source DID is present (if required) 2. Source

Check if path exists.

Ingest a cost data point and check for anomalies. Returns AnomalyResult if anomaly detected, None otherwise.

Terminate a session and commit audit trail. Returns: audit log root summary hash, or None if audit disabled

Validate governance attestation in PR description. Args: pr_body: PR description body text required_sections: List of section tit

Wrap an OpenAI Assistant with governance. .. deprecated:: Use :meth:`wrap` instead:: governed = kernel.wrap(assi

Write data to file.

NewTrustManager creates a TrustManager with the given config.

Create a CA with a registered sponsor.

Create a minimal A2A task dict.

Check if an auth method is allowed for the given MCP server. Args: server_name: Name of the MCP server. auth_method:

Create a new namespace, optionally nested under *parent*. Args: name: Unique namespace name (e.g. "finance.trading").

Create a new chain with a root sponsor.

Get current value of a session attribute.

Get the current trust score for an agent.

Evaluate policy and return a signed governance receipt (fail-closed on signing error).

Check if an agent is currently revoked. Automatically handles expiry of temporary revocations.

Load a Policy document into the engine.

Load all YAML policy files from a directory.

Create a new trace context. Args: operation_name: Name of the root operation Returns:

Process a natural language query. Args: natural_language: The query in plain English user_id: User m

Govern and execute a query call. Runs the full governance pipeline (collection check, rate limit, content scan, audit) through the go

Check content for injection patterns before writing to memory. Returns a ``ValidationResult`` indicating whether the write should pro

Verify trust headers and return *(result, error_body | None)*. *error_body* is a JSON-serialisable dict when verification fails, or `

Wrap a Bedrock Agent Runtime client with governance. Args: client: A ``boto3`` ``bedrock-agent-runtime`` client. Returns

Governance gate around each tool execution. Performs the following checks **before** the tool runs: 1. Tool allowlist / blocklist en

Helper: create a session and register it for cleanup.

Return a :class:`GovernanceRunHooks` backed by this kernel. Pass the returned object to ``Runner.run(hooks=...)``:: kernel = Ope

Process a user message through the governance pipeline. Steps: 1. Rate limit check 2. Policy enforcement (PII, conte

Check if an agent can make a request. Returns True if allowed, raises RateLimitExceeded if not.

Compare current snapshot against baseline.

Create a session token bound to an agent and optional user context. Args: agent_id: Agent identifier that will own the session.

Scan *text* for prompt injection patterns. Args: text: The input text to screen. source: Identifier of the component

Generate a new agent identity with an Ed25519 key pair.

Retrieve a document by ID. Args: document_id: The document ID Returns: The docu

Record token usage for an active allocation. Emits SIGWARN or SIGSTOP as thresholds are crossed.

Register an agent type with the control plane. Args: agent_type: The agent class to register agent_i

Request provisioning of a new agent. If policy requires approval, agent starts in PENDING_APPROVAL. Otherwise, goes directly to PROVI

Revoke this credential immediately. Args: reason: Human-readable reason for revocation.

Generate a new keypair, create a rotation proof, and update the identity. The old public key and rotation proof are stored in key history.

Scan a list of text chunks and return a result per chunk. Args: chunks: List of text strings (document page contents or passages)

Start a new trace. Args: operation_name: Name of the operation metadata: Additional metadata

Create a minimal :class:`AuditEntry` for testing.

Create a PolicyViolation for testing.

Calculate a trust score (0-10) based on capabilities and privacy. The trust score helps clients make informed decisions about agent

Compute exact Shapley values for each agent. φ_i = Σ_{S⊆N\\{i}} [ |S|!(n-|S|-1)! / n! ] * [v(S∪{i}) - v(S)]

Create provenance metadata for data.

Create a signed card from an identity. Args: identity: The identity to create a card for Returns: A

Generate content with governance enforcement. Validates prompt content against the configured AGT manifest at the ``input`` intervent

Intercept direct messages between agents. Checks for: 1. Tool call governance (``FunctionCall`` messages) — allowlist, blo

Parse datetime string. Args: date_string: Date string to parse format: Optional strptime format or p

Reconcile a CRD spec — create, advance, or rollback as needed.

End-to-end scenario evaluation. Args: workspace_root: Tempdir that will host the synthesized governance tree. Must be an exis

Wrap a LlamaIndex query engine, chat engine, or agent with governance. Intercepts: - query() / aquery() - chat() / a

DefaultTrustConfig returns sensible defaults.