Record an interaction and update the trust store.

Record a verdict in history and invoke violation handler.

Internal: record a score and send to Langfuse if live.

Redact sensitive data from text.

Reset the global store between tests.

Resolve a path for both current and legacy monorepo layouts.

Map an ACS :class:`InterventionPointResult` to an AGT :class:`EvaluationResult`.

Render a tiny bundle with stock library + the custom policy and evaluate one query. Returns the verdict body.

Call LlamaFirewall's scan API with graceful error handling.

Write a value to $GITHUB_OUTPUT.

Create two registered agents and a handshake instance.

Convert skill description to action name.

Return the SSN regex from the shared ``PII_PATTERNS`` tuple.

Install a stub verifier that either raises EntraTokenError or returns the supplied claims dict.

Run a synchronous function *iterations* times and return latency stats.

Run a synchronous function *iterations* times and return latency stats.

Check if a security tool is available on PATH without executing it.

Map a toolkit tool_name to a Cedar action identifier. Converts snake_case tool names to PascalCase Cedar actions: ``file_read`` → ``FileRead`

Return a rich style string for a trust level.

Update trust score based on tool usage.

Return a minimal valid AgentIdentity dict for mutation testing.

Raise ValueError if mod_name is not in the governance module allowlist.

Validate path is within sandbox and safe to access.

Write a log entry to the flight recorder.

Write a marketplace policy YAML and return the path.

Add a component with trust identity.

Add a file path to the data graph

Register or update an org policy.

Add a new safety correction based on a failure. Args: task_pattern: Description of the task that failed

Governed asynchronous invocation. Async counterpart of :meth:`invoke` — applies identical pre-/post-execution policy

* Analyze the current state of the agent graph for cascade risk.

Analyze a silent signal event and determine learning strategy. Silent signals are implicit feedback that requires immediate

Atomically apply an EMA reputation update and bump session counters. Implementations MUST hold their internal lock across the read-mo

Governed async streaming chat.

Build the smallest external accountability export from an ``AuditEntry``. The export is an interoperability shape, not a new AGT runtime evidence

Return the curated 100-prompt benchmark dataset.

Check if a principal can access a tool. Args: principal: The principal requesting access. tool_name: Name of the tool

Check if this credential can access a specific resource. If no resource restrictions are set, access is allowed by default. Args:

Encode a dict with deterministic key ordering for signing/hashing.

* Construct a collapsible SLO category item.

Apply policies in priority order on top of the base. Later policies win for simple scalar fields; list fields are unioned.

True iff `agentDid` may use `handleName` for `actionClass`.

Check if user can access matter.

Check if two package versions are in a tested-compatible range. Args: pkg_a: Name of the first package. ver_a: Version of the fir

Check content against blocked patterns. Returns violation reason or None.

Check content against blocked patterns. Returns (allowed, reason).

Detect hardcoded API keys, tokens, and secrets.

Detect external calls without circuit breaker.

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 transition from the last item to *new_item* is rare. Builds a transition frequency map from *sequence* and flags rare tran

Detect security functions that unconditionally return True.

Detect young, low-star projects promoted across multiple orgs. Args: username: GitHub username. target_repo: Optional target repo

Check if metrics meet minimum thresholds. Returns: (passes, reason) tuple

Check if an agent is trusted. Returns the agent's overall trust score, trust level, and all five trust dimensions (competence, integrity, ava

Create a child span (e.g., when spawning a sub-agent or delegating).

Remove expired temporary revocations. Returns: Number of entries removed.

* Clear all logs

Clear all events from the stream (for testing).

Clear all metrics. Useful for testing or reset scenarios.

List all active sessions.

Perform a security review of a file.

Compare expected vs actual output with optional fuzzy tolerance. Args: expected: The golden expected output. actual:

Compare two policy dicts and return drift findings. Missing keys on either side are flagged as *critical* drift. Args: s

Execute a compression cycle: identify old episodes, summarize, and optionally archive. Args: summarizer: Optiona

* Compute the blast radius for a given agent: how many agents would * be affected if this agent fails completely.

Compute SHA-256 hash of codebase content. Args: codebase_content: String representation of codebase Returns:

Return sponsored agent's own score (Public Preview: no sponsor boost).

Compute ring assignment from trust score.