Register an action from a capability manifest.

Score a single message for escalation patterns. Matches against both original and normalized text to catch evasion while preserving r

Set the baseline snapshot for a server.

Stop the processor and flush remaining events.

── SQL ──────────────────────────────────────────────────────────────────────

Return issues matching a given rule id.

Return a mock LlamaIndex engine (no SDK required).

Create a PluginManifest with sensible defaults.

Build a real ``PolicyDocument`` so the agent-os ``PolicyEvaluator`` is happy, and attach the sandbox-provider-specific allowlists as extra fie

Create a policy with optional rulesets for testing.

Append an audit event to the internal audit log. Records the event only when log_all_calls is enabled. Args: event_type:

Trace an outgoing A2A call. Args: target_agent: ID of the agent being called. task: Task description or A2A task ID.

Add a step to the workflow.

Check if agent has required capabilities.

Classify an action and cache the result.

Release all bonds (session termination).

Create and submit an escalation request. If fatigue detection is enabled and the agent has exceeded the threshold, the request is imm

Execute a shell command with safety checks.

Execute a single saga step with timeout and retry support. Cancellation semantics on timeout: ``asyncio.wait_for`` cance

Execute a registered tool. Args: tool_id_or_name: Tool ID or name parameters: Tool parameters

Generate a new signed pre-key and sign it with the Ed25519 identity key.

Scan a tool response for threats before it reaches the LLM. Runs the ``MCPResponseScanner`` against the response content and applies

Check if a card is verified (with caching). Uses TTL-based caching to avoid repeated verification. Returns False if the agen

Log a verification operation. Args: operation: Type of operation (e.g., "verify_embeddings") inputs: Input d

Observe a signal without blocking. This is the entry point for signals. The signal is queued and will be processed a

Read file contents.

Record an evaluation result from the judge.

Record a message and return the current loop score. Returns: Loop score in [0, 1]. Higher = more likely in a dangerous loop.

Record that *agent_id* consumed *tokens* and *cost_usd*.

Register an identity.

Create a signed message envelope. Args: payload: Serialized MCP payload to sign. sender_id: Optional sender identifie

Send a signal to this agent. Returns True if signal was delivered, False if blocked/pending.

Calculate square root.

Verify an Ed25519 signature against a public key. Args: public_key: Raw Ed25519 public key bytes. data: The original

GetEntries returns entries matching the given filter.

Create Ed25519-Timestamp auth header for prekey upload.

Create a mock ToolCallHookContext.

Return a minimal valid policy as a plain dict.

Add an agent DID to a namespace. Args: namespace_name: Name of the target namespace. agent_did: DID of the agent to a

Add a node to the graph

Add a conversation turn to the history. This implements the FIFO Sliding Window: - If we have < max_turns, just add

Assign full liability to the direct-cause agent.

Characteristic function v(S): value (fault contribution) of a coalition. A coalition's value is the sum of weighted fault contributions of it

Check and consume a token for the given agent/action pair.

Audit agent security configuration.

Compute deterministic policy hash.

Create a context for an agent in user space. Allocates all kernel resources the agent needs — a signal dispatcher, VFS instance, and

Create an isolated scope for a session. Args: session_id: Unique session identifier. agent_did: DID of the agent owni

Generate a self-signed X.509 certificate from the agent identity. Uses the agent's Ed25519 identity key for certificate signing (RFC

Evaluate a Rego query against input data. Args: query: Rego query (e.g., "data.agentmesh.allow") input_data:

Evaluate current SLO status.

Evaluate whether a tool call should be allowed.

* Execute an action through the full governance pipeline: * 1. Evaluate policy * 2. Check trust score * 3. Log to audit trail

Freeze the policy engine, preventing further mutations. After calling ``freeze()``, any attempt to call ``add_constraint()``, ``set_a

Load a policy bundle from a YAML file.

Generate an Ed25519 keypair and store it in memory. Args: agent_id: Unique identifier of the agent. Returns:

Get or create a trust score for an agent.

Get the reputation score for an agent. Args: agent_id: Agent identifier Returns: ReputationScore if

Check if the agent has a specific capability. Checks the deny list first, then searches for a valid, matching grant. Args:

Load the revocation list from a JSON file. Args: path: File path to read the JSON data from.

Load and validate a profile from a YAML file. Raises ``ProfileError`` with a path-prefixed message on any schema violation, missing key, wron

Merge governance documents into a flat priority-sorted rule list. Args: documents: Parsed governance documents in **root-first** order.

Inject a prompt injection attack against a target.

Return the SHA-256 hash of a raw Ed25519 public key as lowercase hex.

Record a new violation and return the record.

Record a failed interaction with an agent.

Reset with new data. Frame counter resets.

Score a message for offensive intent patterns. Matches against both original and normalized text to catch evasion while preserving re

Sign arbitrary data and return the signature bytes.