Merge multiple policies using most-restrictive-wins semantics. - ``max_tokens``, ``max_tool_calls``, ``max_concurrent``, ``timeout_seconds``:

Compute a stable fingerprint from merge keys.

Compute the SHA-256 hash of this entry's canonical fields. Returns: Hex-encoded SHA-256 digest.

Create a new attestation record (used by Control Plane). Args: agent_id: Agent identifier codebase_hash: SHA

Create execution context for an agent. The policy is **deep-copied** so that the session is pinned to the policy that was active when

Helper: Create a log stream signal.

Discover governance policy files from action path up to root. Walks the directory tree from ``action_path`` upward to ``root``, collecting ``

Apply enforcement for one intervention-point result. In ``enforce`` mode a ``deny`` raises :class:`AgentControlBlocked`, and an ``esc

Export an SLO evaluation as a Phoenix EVALUATOR span.

Mock filter implementation matching GovernanceHubScript

Content hash for change detection.

Get a message by ID. Args: message_id: Message ID Returns: Persisted message or

Get the callable function for a tool. This returns the function object but does NOT execute it. The caller (Agent Runtime) is respons

Return a snapshot of current usage for *agent_id*.

Execute a tool call through governance. Returns the mock result string on success, or None if blocked.

Handle a failure in agent communication. This is the main entry point for failure recovery. It: 1. Diagnoses the failure usi

Check transparency requirements before allowing a tool call.

Load an MCP configuration from JSON or YAML. Args: path: Path to a JSON/YAML config file. Returns: Parsed config object.

Load runbook definitions from a YAML file. Validates the document structure and rejects entries that would silently corrupt the registry — em

Check if this grant satisfies a requested capability. Uses simple startswith matching for scope checking.

Filter *result*, redacting sensitive content in-place. Accepts an ``ExecutionResult`` (from ``agent_os.stateless``) or any object wit

Execute a Face→Hands pipeline with kernel-level validation. 1. Face agent produces an ExecutionPlan from the task 2. Kernel validates plan ca

Calculate penalty from a list of policy violations. This is a simple helper for computing penalties outside of the full PolicyReward cla

Evaluate an action against governance policy. Returns GovernanceResult with allowed=True (pass-through) or allowed=False (blocked wit

Raw Ed25519 public key bytes (32 bytes).

Query audit entries with optional filters.

Read all events from the stream.

Record that *alert* was sent (update window timestamp). See :meth:`consume` for the atomic check-and-record alternative.

Record a tool call. Args: tool_name: Name of the tool. latency_ms: Call latency in milliseconds. success:

Record a task event for an agent. Returns False if agent not found.

Redact all PII from text.

Register an agent in the marketplace.

Resolve an agent's sigma. Returns 0.50 default when no scorer is configured.

Evaluate an action against the loaded governance policy. Returns a dict with ``decision`` ("allow" | "deny" | "audit"), ``reason``, a

Compute risk assessment for a discovered agent.

Set the context data for an agent for ABAC evaluation. Args: agent_role: The agent role/ID context: Dictiona

Set a gauge metric value.

Maximum security policy: deny all except allowlist. Suitable for high-risk environments where every action must be tightly controlled

Suspend this sponsor (and all their agents should be suspended too). Args: reason: Optional human-readable reason for suspension.

Like check(), but returns False instead of raising.

Run health check on the inactive environment. Args: health_check_fn: A callable that returns True if the environment is healthy.

Deterministic policy check against all registered policies. Args: action: Dict with at least ``tool`` (str) and ``arguments`` (di

Verify a handoff from source agent to target agent. Checks: 1. Source agent meets trust threshold 2. Target agent me

Always valid.

Verify the Ed25519 signature of a plugin manifest. Args: manifest: The manifest to verify (must include a ``signature``). public_

Wrap an Anthropic client with governance. .. deprecated:: Use :meth:`as_message_hook` instead for a non-invasive inte

Wrap an OpenAI Assistant with governance. This is the primary wrapping method, consistent with all other adapters. OpenAI Assistants

SetRingPermissions replaces the allowed actions for a given ring.

Transition moves the agent to a new state if the transition is valid.

Format a datetime for display, handling None.

Centralized error handler.

Create a mock AgentContext.

Create a mock CrewAI crew.

Create a mock FunctionInvocationContext.

Create an MCPGateway with a default policy.

Helper to create a valid DelegationLink.

Create an AgentTrustMiddleware wrapping a dummy get_response.

Create a mock LangChain chain/agent.

Create a mock smolagents step.

Return a MeshTracer backed by an in-memory span exporter.

Return rules whose condition matches the given field/value pair.

Print a formatted results table.

Create, record, and surface a PolicyViolationError. Appends the error to the violations list and calls on_violation. Args:

Record call for audit.

Remove dangerous env vars that could escape sandbox hardening.

Map a 0-100 score to a letter grade. Scans ``GRADE_THRESHOLD_LIST`` top-down (highest threshold first) and returns the first letter whose thr

Start a local HTTP server serving *data* on GET /. Returns (server, url).