Build a context dict for PolicyEvaluator/CedarBackend. Constructs the principal/action/resource triple that Cedar and OPA backends ex

Build: A → B → C

Flatten action + params into a single searchable string. Walks both keys and values recursively. Previously this only iterated `param

Build a 2-link chain where the second link has the given expiry.

Check text length.

Return True if *module_name* is importable.

Auto-detect the strongest available OCI runtime.

Discover an advanced provider via entry_points.

Block the call when the v4 tool-call/token budget is spent. A governed ``query``/``chat`` is a budgeted operation, so

Execute an action through the kernel with policy checks. This is the primary method for agents to perform actions. All actions are:

Best-effort agent name extraction.

Return 1-based line number of first match, or 0.

Flush pending writes to database.

Recursively convert nested mappings/lists to hashable tuples.

Return a context dict that satisfies every built-in gate.

Return demo peer data for display when no live data is available.

Best-effort mapping from metric name to anomaly type.

Add the authorization fingerprint to checkpoint metadata.

Return True if the account shows strong organic credibility.

Return whether a runtime module section is enabled.

Validate that a string matches the expected did:key format.

Helper: return a LangChainKernel (concrete subclass) with given policy.

Create a mock CrewAI agent with optional tools and memory.

Create a mock CrewAI crew.

Return a mock ``GenerativeModel`` that responds with ``response``.

Create a mock LangChain-style tool with _run and _arun.

Return a fake urlopen that walks through `side_effects` on each call. Each side effect is either an Exception (raised) or a dict (wrapped int

Map LlamaFirewall string verdict to FirewallVerdict.

Create a mock that works as a context manager returning JSON data.

Evaluate the AGT input intervention point and apply transforms.

Append a timestamped audit event to the internal log. Args: event_type: Short label (e.g. ``"agent_start"``, ``"tool_end"``).

Redact a sensitive value for safe logging. Replaces sensitive data with a stable hash-based identifier. The original data never appears in ou

Register an agent through the public API; returns (sk, did).

Save intent with optional optimistic concurrency check.

Deterministic but fresh-looking data on each run.

Convert a version string to a comparable tuple. Defensive against versions that slipped past :func:`PluginManifest.validate_version` — for ex

Compute HMAC over peer's critical fields. See class docstring on `_peer_hmac_key` — this HMAC is an in-process integrity check, not p

Simulate on_llm_start → on_llm_end (or on_llm_error).

Normalize an :meth:`SandboxClient.exec` response. The 0.1.0b1 SDK returns a typed result object whose example usage is ``out.stdout`` / ``out

Check if version is within [min_v, max_v] inclusive.

Generate or return a cached in-memory Ed25519 key.

Register a trace for an agent.

Add context labelled as *cold*.

Add evaluation feedback to a run. Args: run_id: Run ID to attach feedback to key: Feedback key (e.g. "correctness", "

Advance a rollout to the next step.

Post-execution governance check for a tool call. Inspects tool output for blocked patterns via the AGT ``output`` interventi

Analyze a manifest provided as a dictionary.

Apply a reputation event to update the score. Args: event: The reputation event to apply

* Build a tooltip string for the governance mode status bar item. * * @param mode - The active governance mode. * @param activePolicies - Number of

Build an execution graph from a trace.

Build a system prompt with prioritized context.

Return the :class:`SnapshotBuilder` bound to ``ctx``. The builder is created on first access and reused across intervention points. H

RFC 8785 JCS canonical JSON (signature fields excluded). Includes a per-receipt nonce to prevent replay attacks: even if two receipts

Return the deterministic JSON string that was signed.

Return whether *module_name* is permitted under the sandbox policy. This is a **policy predicate** — it answers "is this module on the

Check if a tool is allowed by policy.

Check if an agent meets the trust threshold. Returns: Tuple of (meets_threshold, trust_score)

Check if package name is suspiciously similar to a popular package.

Remove all extractors. Primarily for tests.

Compile a profile to a Cedar policy string. The output is deterministic and ordered to match the source profile.

Compile a profile to a Rego policy string. Output is deterministic and ordered to match the source profile.

Compute a cache key for content. Args: content: The content to hash metadata: Optional metadata to i

Compute ADR 0010 canonical report-data binding as lowercase hex.

Percentage of controls covered.

Create a new replay session for an agent. Args: agent_id: Agent identifier start_time: Start of repl

Mark a plugin version as deprecated. Args: name: Plugin name. version: Plugin version. reason: Human-read