Save information to kernel memory with governance checks. The text content is validated at the AGT ``input`` intervention point befor

Return True if the TTL has elapsed since the last rotation. Returns: True if the elapsed time exceeds ``rotation_ttl_seconds``.

Normalize Shapley values to sum to 1.0 (liability shares).

Register a handler for *signal*.

Called after a tool completes execution. Governance actions performed: 1. **Output filter** — scans the tool's result against

Called when a tool finishes executing.

Called when a tool errors out.

Verify agent trust before tool execution.

Called when a tool starts executing.

Compute the Pareto-optimal set of models for a task. A model is Pareto-optimal if no other model is both cheaper AND higher quality w

* Parse natural language task description into structured agent spec

Log a policy decision at INFO level.

Create an InputGuardrail that enforces governance policies. Checks input content against blocked patterns and tool call limits.

Run post-execution validation and return a structured result. Args: ctx: Execution context for the governed operation.

Snapshot for the ``post_tool_call`` intervention point (§2.6).

Helper to print a context object in a readable format.

Print a section header.

Print a section header.

Print a section header.

Process a raw signal from any source. Args: raw_signal: Raw input signal (can be text, file change, log, etc.)

Process a refund request with governance checks.

Purge redundant lessons from the wisdom database. Args: audit_results: Results from audit_wisdom_database()

Log a quarantine request (Public Preview: no enforcement).

Query persisted alerts.

Get all audit entries of a specific type.

Read and parse JSON file.

Read events that haven't been processed yet. If last_processed_timestamp is None, returns all events.

Recalculate risk score based on current signals. Called every ≤30s or immediately on critical signals. Only signals from the last 24

Reconstruct BOMs for all decisions by an agent in a time range. Args: agent_id: The agent to query decisions for. sta

Increment the audit-entries counter. Args: event_type: Type of audit event.

Increment the audit-events counter. Args: event_type: Type of audit event (e.g. ``"trust_update"``).

Record a policy check result.

Record a disagreement between two models.

Record a crash.

Record a scope chain depth.

Penalize an agent for a failed action. Args: agent: The agent to penalize. Its ``trust_score`` is modified in pla

Record a task failure. Returns new trust score.

Record a failure event (timeout, error, etc.). Args: agent_id: Agent that failed failure_type: Type of failu

Record the start of an operation

Record output quality signal from downstream consumer.

Record recovery after SIGKILL.

Record a successful action, boosting specified dimensions.

Record a successful task completion. Returns new trust score.

Increment the ``tool_call_count`` budget by ``count`` (default 1). Hosts call this after a ``post_tool_call`` returns successfully, m

Update agent states based on heartbeats and success rates.

Register an agent with both AgentMesh DID and Entra Agent ID.

Register an executor. Args: executor: The executor instance plugin_id: Optional ID actio

Register a signal normalizer for a specific signal type. Args: signal_type: Type of signal this normalizer handl

Register an MCP tool. Args: tool_name: Name of the tool tool_info: Tool metadata (description, input

Reload policies from *directory* with an exclusive write lock. Acquires the write side of the RW lock so that no evaluations can proc

Remove a package from tracking.

Reply to a request message. Args: original_message: The original request message payload: Reply payl

Raise if the key handle is not TEE-bound. Intended for use in handshake or policy enforcement layers (PR 4/5). Raises: KeyAcquisitio

Reset global config (useful for testing).

Reset daily budgets (call at start of each day).

Check whether *agent_id* may access *resource*. Args: agent_id: The requesting agent's identifier. resource: Name of

Return trust headers to attach to outgoing responses.

Restore the session's VM memory from a previous snapshot.

Govern and execute a retrieve call. For LlamaIndex ``BaseRetriever`` — delegates to ``.retrieve()`` instead of ``.query()``.

Revoke a single session token. Args: session_token: Session token to revoke. Returns: ``True`` when the sess

Rotate credential if it's expiring soon. Args: credential_id: ID of the credential to check. Returns: The ro

Route based on current trust score.