Get violations that block execution (CRITICAL or HIGH).

Look up agent by AgentMesh DID.

Walk the causal graph from *episode_id*. Parameters ---------- direction : "backward" | "forward" | "both"

Resolve the config path from args or AGENTOS_CONFIG env var.

Get identity credentials for MCP authentication.

Compute current severity based on progressive config and execution history.

Get statistics about the dataset

Resolve effective policy for an organization (base + org overrides). When *organization* is ``None`` or the org has no overrides, the base

Get an intent by ID, or None if not found.

Return a copy of the interaction audit log.

Return a namespace by name. Args: name: The namespace name to look up. Returns: The matching AgentNamespace.

Main method: Get prioritized context for a query. Args: query: User's query/task global_wisdom: Base

Retrieve the quality profile for a specific plugin version.

Get current queue size for a topic. Args: topic: The topic to check Returns: Nu

Retrieve the global profiling report.

Classify drift severity based on profile thresholds. Returns: "pass": Within threshold "warning": 1-1.5x thr

Get rate limit stats for an agent.

Return aggregate statistics. Thread-safe.

Get statistics about learned patterns.

Get the lifecycle status of a plugin version.

Get the current system instructions as a prompt.

Return the capability configuration for a tier. Args: tier: Tier name (must be a key in :data:`TRUST_TIERS`). Returns: The :

Acquire a token from the Azure IMDS endpoint.

Get the HF token from config or environment.

Get a trace by ID

Get current traffic split between old and new versions. Returns: (old_percentage, new_percentage) tuple

Get the trust level based on reputation score. Args: agent_id: Agent identifier Returns: Appropriat

Convert reputation score to trust level. Returns: Appropriate TrustLevel based on current score

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

Return an async tool callable guarded by pre/post tool-call intervention points.

Handle MCP initialize request.

Handle MCP tools/list request.

Handle an MCP tool invocation.

Simulate identity spoofing attack.

Import a batch of evaluations.

Inject session state into a policy evaluation context. Adds ``session.*`` keys so policy conditions can reference them: ``session.dat

Create a concrete experiment from a template.

Check if delegation is active and not expired.

Check if an agent is currently hibernated

Check if a tool is in the allowed list after canonicalization. Both the tool name and all entries in the allowed list are canonicaliz

Check whether a run has been cancelled. Args: invocation_id: The ADK invocation ID to check. Returns: True i

Check if an agent is registered.

Serialize as deterministic JSON compatible with RFC 8785-style ordering.

Lint a file or directory of YAML policy files.

List all loaded policy names.

List all registered agents, optionally filtered by status.

Load a plugin manifest from a YAML file. Args: path: Path to the manifest file or directory containing one. Returns: Parsed

Load a plugin manifest from a YAML file. Args: path: Path to the manifest file or directory containing one. Returns: Parsed

Load a marketplace policy from a YAML file. Args: path: Path to the policy YAML file. Returns: Parsed MarketplacePolicy.

Load a Policy from a YAML file path, YAML string, or dict.

Load all SLO specs from YAML files in a directory.

Parse and register a policy from a YAML file. Resolves ``extends`` references recursively with additive-only merge semantics. Parent

Log an agent action.

Log an evaluation with SLO scores. Args: trace_id: Unique trace identifier agent_id: Agent identifier slo

Log a policy enforcement decision.

Log an outgoing request.

Compute the deterministic manifest hash used as the signature message. Timestamps and mutable scoring fields are excluded so the hash is sta

Check if score meets a threshold.

Check if score meets a required threshold.

Bare minimum governance: just a basic token cap. The lightest possible policy — only a token limit is enforced. All other settings us

Batch-monitor all active sessions with early exits. Skips archived/terminated sessions via the active index and skips health

Normalize text to defeat common evasion techniques. Handles: unicode homoglyphs, leetspeak, zero-width characters, excessive whitespace, comb

Observe a histogram metric. Args: name: Metric name value: Observed value labels: Metric

* Register a callback for transport disconnect events. * * `reason` is `"client"` for caller-initiated `disconnect()`, `"server"` * for relay

Called when an agent begins execution. Governance actions performed: 1. **Content filter** — scans available input text against

Content-filter the raw user message. Validates the ``user_message`` structure defensively: if ``parts`` is not iterable or individual

Print a formatted header.

Print a kernel log message

Print a formatted section header