Commit audit trail and return hash chain root (None if audit disabled).

Drop agent entries whose deques are entirely outside the window. Caller must hold self._lock.

Compare two versions of the same tool.

Compile pattern strings from ``self._injection_config`` into ``re.Pattern`` objects stored on the instance. Detection methods iterate

Compile pattern strings from ``self._config`` into ``re.Pattern`` objects stored on the instance. Detection methods iterate these inst

Compute an aggregate value over filtered records.

Compute how complete the BOM is (0.0-1.0). Based on how many required fields are present.

Compute SHA256 hash for Merkle chain.

Compute deterministic hash of manifest.

Compute summary statistics for the graph.

Return ``True`` if every condition key/value is present in *context*.

Build human-readable config diff message.

Determine severity of a config difference.

Configure TLS verification for remote MCP endpoints.

Consume messages from Kafka.

Recursively check whether ``value`` contains an approval key. Catches Cyrillic-confusable ``approvеd``, case variants ``Approved`` and ``APPR

Check if parameters might contain PII

Convert an audit log entry to a ReplayEvent

Convert JSON Schema to Gemini Schema proto.

Count violations by severity

Count violations by type

Count entries within the rolling window, pruning expired entries.

Count logs by level

Count operations in the last N seconds

Count checked boxes in a specific section. Args: section_title: Title of the section to find (e.g., "1) Security review") pr_body

Create and start the HTTP server.

Create the WebSocket server.

Create this agent card for discovery.

Create an LLM client, auto-detecting backend from env vars.

Create an LLM client, auto-detecting backend from env vars. Returns: (client, backend_name) tuple.

Create an LLM client, auto-detecting backend from env vars. Returns (client, backend_name). Falls back to None with BACKEND_NONE when no

Create an incident from multiple correlated signals. Groups signals into a single incident, using the highest severity among the sign

Create default execution context based on action type

Create Python script for container execution. Args: func_source: Source code of the function. func_name: Name

Create an incident from a single signal.

Create a natural language query describing the change.

Create query from log entry.

Create RAG-style prompt

Create a new replica for an agent type

Create a new rollout from spec.

Create a rotation proof: the old key signs the new public key.

Create a trust-verified wrapper for a function.

Get the current OTEL span if one is active.

Decode the header section of a JWT.

Build the default certification criteria for all tiers.

Get default (minimal) permissions for an agent

Simple rule-based response (replace with LLM in production).

Default summarization strategy: extract common patterns. Args: episodes: List of episodes R

Log a policy violation at ERROR level. This is the default handler used when no custom ``on_violation`` callback is provided.

Deliver alert to a single channel.

Push all pending messages to a newly connected agent. Messages stay in the inbox until the recipient explicitly sends an ``ack`` fram

Wrap ``invoke`` calls to detect and govern sub-agent delegation. Monitors the agent's ``invoke`` method (if present on the original o

Detect attack from call trace

Detect which LLM backend to use from environment variables.

Detect which LLM backend to use from environment variables.

Detect which LLM backend to use from environment variables.

Detect if two sections conflict. This is a simple heuristic-based detection. In a production system, you might use N

Detect when a CrewAI agent delegates work to another agent. Wraps the ``delegate_work`` or ``execute_task`` related delegation method

Core detection logic — runs all check methods and aggregates.

Simple language detection based on syntax.

Detect PHI patterns in text.

Auto-detect signal type from raw signal structure. Args: raw_signal: Raw signal data Return

Embedding-based detection. Simplified version - in production would use real embeddings and cosine similarity.

Pattern-based jailbreak detection

Determine the appropriate recovery strategy. Args: diagnosis: Failure diagnosis from scak manifest: Agent ca

Determine the content tier for a section. Args: section: The section to classify doc_type: Type of d

Perform a DH ratchet step with a new remote public key.

Return the closest match from *candidates*, or ``None``.

Normalize a hex-encoded SHA-256 digest to lowercase and validate length.

Fan out a batch to all registered sinks with error isolation.

Dispatch execution to appropriate handler Note: This returns simulated execution results for demonstration purposes. In prod

Execute the core handshake: generate nonce, verify it comes back.

Extract text from a document object.

Pop up to max_batch_size events from the queue.

CMVK drift-based consensus. Algorithm: 1. Convert responses to vectors (embeddings or numeric) 2. Calculate

Drop the oldest BACKGROUND priority message from the queue. Args: topic: The topic to drop from