Create a configured SSL context for mTLS. Args: server_side: If True, create a server-side context that requests

Get all criteria applicable to a tier (cumulative).

Simulate circular dependency deadlock between agents.

Begin decommissioning the agent.

Delete a file from the file system. Args: path: Path of the file to delete agent_id: ID of the agent

Add a capability to the deny list. Denied capabilities take precedence over any matching grants. Args: capability: Capab

Deserialize from bytes.

Detect PHI patterns (medical record numbers, diagnosis codes).

Find out-of-sync versions among components. Expects a dict mapping component names to version strings. When two or more distinct ver

Gracefully close the WebSocket connection.

End a run. Args: run_id: Run ID to end outputs: Optional output data error: Optional error message

End the current span and pop from stack.

End a trace

Run all policy checks for an action. Returns first violation or allowed.

Raise when *code* contains denied subprocess execution patterns.

Enrich episode metadata with tags and length metrics.

Pull *image* if it is not already present locally. Authentication is resolved through Docker's standard credential chain (``docker lo

Canonical byte string that the requester signs to authorize an escrow. Both ProofOfOutcome.create_escrow() and callers must use this functio

Evaluate an authorization request against an external trust artifact. Args: agent: The agent requesting task authorization.

Evaluate this condition against a context dictionary. Args: context: Runtime context with values accessible via d

Evaluate using pattern matching.

Evaluate using LLM.

Run *vectors* (defaults to built-ins) and return a report.

Check whether *agent_id* may access data described by *data_label*.

Evaluate a policy check with escalation support. If the policy would deny due to ``require_human_approval``, this creates an escalati

Check if a metric value passes this criterion.

Evaluate multiple contexts concurrently. Each context is evaluated in its own asyncio task. All tasks share the same concurrency con

Public access to the AGT ``input`` intervention point evaluation.

AGT ``output`` evaluation for buffered LangChain output.

Execute this step synchronously.

Execute a task - ALL actions go through the kernel.

Execute a task with trust verification for any delegation. Verifies this agent's own trust score before execution and records the int

Export inventory as JSON.

Export all reputation data for network-wide propagation. Returns: Dictionary containing all reputation scores

Export SLO evaluation as Datadog metrics. Creates metrics for: - agent_sre.slo.status - agent_sre.slo.budget_remaining

Extract agent DID from certificate subject or SAN. Looks for a URI SAN matching ``did:mesh:*``, then falls back to the subject SERIAL

Extract potential search tags from *text*.

Mark task as failed.

Check if an action path matches a policy's scope glob. Args: policy_path: Path to the governance.yaml file. scope_pattern: Glob p

Return all credential-like matches found in a string. Args: value: String content to inspect. Returns: A lis

Format a citation as a human-readable string. Args: citation: Citation to format Returns:

Format scan results as a Markdown report.

Format a reputation report for display.

Create an AgentIdentity from a JWK.

Import from a JWK Set.

Create an evaluator from a YAML config file.

Deserialise from a dict.

Create from dictionary.

Convert an Ed25519 key pair to X25519 via birational map. Args: ed25519_private: 64-byte Ed25519 secret key. ed25519_

Parse session_attributes from a policy YAML string. Example YAML:: session_attributes: - name: data_sensitivity

Parse a rate limit string like '10/minute'. Args: rate_string: String in format 'N/period' where period is

Generate searchable tags from an episode.

Generate a warning message for risky requests that aren't blocked. Returns None if no warnings are needed.

Get sponsor by ID. Args: sponsor_id: Unique sponsor identifier. Returns: The HumanSponsor, or None if not fo

Intercept checkpoint reads — validate fingerprint before resume.

* Get active policies

* Get policy by ID

* Get the effective blast radius policy for a trust tier.

* Get approval request by ID.

* Get logs for today

Get currently firing alerts

Get an agent by ID.

Get all ledger entries for an agent.

Get current trust score for an agent.

Return the full audit trail.

List available tools.

Return the quality badge for a plugin, defaulting to UNRATED.

Return the bucket state stored for *agent_id*, if present.

Get sponsor by email. Args: email: The sponsor's email address. Returns: The HumanSponsor, or None if not fo

Return the number of calls inside the active window. Args: agent_id: Agent identifier to inspect. Returns: T

Return items labelled *cold*, optionally filtered by *query*.