Detect templated repo creation: many repos with same suffix in a short window.

Scan a string for policy-blocked patterns. Args: content: The text to scan. Returns: Tuple of (allowed: bool

Check if an instance comes from a forbidden dependency

Validate a tool call against the active governance policy. Raises :class:`PolicyViolationError` if the tool is not allowed or if its

Check value is within bounds.

A coding agent that: - allows file_read on any file - denies file_write to env-y paths (secrets, .env) - escalates rm -rf-style commands

Merge results from both scanners based on the operating mode.

Shortcut: create a detector with custom config and scan.

Ensure storage directory exists and index is loaded.

Extract a stable name for *agent*, falling back to ``id()``. Args: agent: Any object with an optional ``name`` attribute.

Helper to perform GET request.

Return a cached token if still valid, else ``None``.

Extract invocation_id from an InvocationContext or CallbackContext.

Get a managed session or raise 404.

Read a Django setting with a fallback default.

Get current UTC timestamp in ISO 8601 format.

Check if grade is below minimum.

Handle a tools/call request - enforce policy. Args: message: The JSON-RPC message Returns: Modified

SHA-256 hash of a file's contents.

Check if a finding is exempted. Matches exemptions against findings using three strategies (in order): 1. Exact match on category + f

Make an LLM call, dispatching to the correct backend. Returns a normalized response. On API error or missing backend, returns a simulated

Return the ``before_tool_call`` governance function. Returns ------- callable A function conforming to CrewAI's `

Return a mock LlamaIndex engine (no SDK required).

Return ``GovernanceMetrics`` bound to a fresh isolated registry.

Return a ``GovernanceTracer`` backed by an in-memory span exporter.

Create a concrete BaseIntegration subclass for testing.

Create a PluginManifest with sensible defaults.

Create a mock MAF Message with a .text property.

Create a mock model response.

Create a mock OpenAI Runner class with an async run method.

Return the policy's network allowlist as a list of host patterns.

Return the policy's sandbox egress default ('allow' or 'deny'). Reads ``policy.defaults.network_default``. **Falls back to 'deny'** (fail-clo

Create a new execution context for a request. Args: **extra_metadata: Additional metadata to include Returns:

Handle a failed call

Parse Cedar permit/forbid statements from policy content. Returns dicts with: effect, action_constraint, has_principal_constraint, has_resour

Parse a version string into a comparable tuple.

Calculate per-minute rate within the rolling window.

Record a decision and update the hash chain.

Replace potential secrets in text with [REDACTED].

Register an agent and return the response.

Execute one snippet and assert that its outcome matches the article.

Neutralize CR/LF/tab in attacker-controlled fields before they reach ``logger.exception``; prevents log forgery against line- oriented log shi

Save review queue to file.

Scan code blocks within a markdown file.

Map score to letter grade.

Create and sign a plugin manifest.

Start a local HTTP server with canned responses.

Validate and apply a state transition.

Raise if the config-provided env would hijack the child runtime. Suppressed when ``--unsafe-allow-all-commands`` is engaged because in that m

Reject untrusted config-provided working directories in live mode. Even with a trusted launch executable, running it in an attacker-chosen di

Project policy fields onto a :class:`SandboxConfig`. Mirrors :func:`agent_sandbox.docker_sandbox_provider.docker_config_from_policy` but only

Transition to active execution phase.

Track a new agent sponsored by this human. Args: agent_did: DID of the agent to add.

Add a compliance report to the dossier. Args: report: A ``ComplianceReport`` from the compliance engine.

Add a database table to the data graph

Add a compliance requirement. Args: framework: Regulatory framework title: Short description

Add a supervisor agent to monitor worker agents

* Allow a specific violation once

Check if the policy allows access. Args: principal: The principal requesting access. tool_name: Name of the tool bein