Record a task outcome and return updated score. Args: agent_id: The agent identifier. outcome: "success" or "failure"

Record an action for the given agent.

Record a lifecycle event and update state.

Record a trust event and adjust the plugin's score. Args: plugin_name: Plugin identifier. event: Short description of

Record an execution and save state. Args: version: "old" or "new" success: True if task completed su

Record a positive signal (prevents decay, small score bump).

Record a successful interaction, boosting trust.

Record a successful transaction to improve reputation. Args: agent_id: Agent that performed successfully tra

Record task success/failure and update trust score.

Record a policy violation.

Register an adapter class under the given name. Raises: ValueError: If *name* is already registered. TypeError: If *a

Register a runbook.

Register an agent type for scaling

Register an SLO for monitoring.

Register a supervisor at a given level. Args: name: Unique supervisor name. level: Hierarchy level (0 = root, higher

Register a new tool in the registry. Args: name: Tool name (must be unique) description: What the to

Register a workflow definition.

Remove an entry from the DLQ. Args: message_id: Message ID to remove Returns: T

Send a request and wait for a response. This implements the request-response pattern for cases where you need to wai

Request human approval for an action. If the action doesn't require approval, returns auto-approved. Otherwise, creates an escalation

Research and academic policy with generous limits. Allows broad exploration but blocks destructive operations. Suitable for research

Check or wait for a resolution. For ``InMemoryApprovalQueue``, this blocks up to ``timeout_seconds``. For other backends, this polls

Run the CLI with the given arguments and return the exit code.

Evaluate pre_model_call/post_model_call around a single model request.

Execute all steps in a playbook and score the results.

Scan content for secrets.

Scan a directory recursively for dependency manifests. Walks ``directory`` and every subdirectory, picking up ``requirements*.txt``,

Return an anomaly score based on entropy deviation. * Score > 0 indicates anomalous entropy (too low or too high). * Returns 0.0 when

Search inventory with filters.

Set the input/output contract for the agent.

Check if we should retry on this exception. Args: exception: The exception that occurred. Returns: True if s

Sign the action (simulated IATP)

Sign an artifact file and return a :class:`SignatureBundle`.

Sign a data handling policy to receive access. The receiving agent must sign the policy before getting the decryptio

Send signal to a governed agent.

Send signal to agent.

SIGKILL — terminate all kernel operations immediately. Once killed, any in-flight or future invocations will raise ``ExecutionKilledE

Store trust score data for an agent. Args: agent_did: The agent's DID identifier. score: Trust score dictionary (e.g.

Map Entra group memberships to AGT capabilities. Only groups present in ``group_scope_map`` contribute capabilities. Unknown groups are

Advance one frame. Returns true when settled.

Serialize this entry as a CloudEvents v1.0 JSON envelope.

Serialize policy to a dictionary.

Convert to dictionary.

Convert to dictionary for JSON output.

Return a flat dict of prefixed keys for merging into the action context. Only sub-contexts that are not None are included so that missing

Export as SPDX 2.3 JSON.

Context manager that wraps a policy evaluation with an OTel span. Yields a dict that the caller populates with results (rule, action, etc.).

Update an existing file. Args: path: Path of the file to update content: New content age

Validate an agent attestation record. This ensures the agent is running verified code and not a hacked version. Args:

Validate a bundle and return a list of error messages (empty = valid).

Validate handshake compatibility between agents. This checks if the remote agent's capabilities meet the local agent's requi

Validate a policy YAML document against its declared schema. Checks for required fields, valid values, and structural correctness. Args:

Validate an SLO spec and return any errors found. Checks: - Target is between 0-100% - Window is a positive duration - SL

Validate a session token without trusting a caller-asserted agent ID.

Verify agent identity via AVP verification endpoint. Calls GET /v1/agents/verify/{name} and checks the ``verified`` field. Falls back

Complete an intent and produce a verification report. Compares planned actions against what was actually executed. Args:

Verify a peer agent before IATP handshake. This is the core viral mechanism - unverified peers get directed to regis

Wrap a smolagents agent with governance. .. deprecated:: Use :meth:`as_step_callback` with ``step_callbacks=`` instead

CapabilityKillSwitchScope returns the kill switch scope for a specific capability.

Evaluate evaluates the given context using the configured Cedar mode.

Extract runs all registered extractors against context in place. For each registered key whose value is a map[string]any, the extractor is called and

FromJSON deserialises an identity from the public-only JSON format produced by ToJSON. The resulting identity has no private key, so it can Verify sig

-- Admin surface ---------------------------------------------------------- Put stores or replaces a credential.

Use appends middleware to the stack.

Return env keys from ``env`` that would hijack the launched runtime. Compared case-insensitively against ``_COMMAND_RESOLUTION_ENV_KEYS`` and

Build a structured, actionable error message for elevation failures.