Record a standalone policy violation. Args: adapter: Adapter name.

Register a health check for a tool. Args: tool_name: Name of the tool. check: HealthCheck instance, callable, or URL

Register an agent with its factory function for recovery

Register a pipeline component with its identity.

Register an SLI instance for a specific agent.

Register or update the capability profile for an agent.

Register a tool with a cryptographic fingerprint. If already registered, updates last_seen and increments version only when the defin

Release decryption key after policy is signed. Only releases key if: 1. Requester is the intended receiver 2

Unregister the PagerDuty channel from the AlertManager.

Replay a trace, using captured outputs as mock responses. Args: trace: The trace to replay. overrides: Optional dict

Replay the last N minutes of an agent's life exactly as it happened. This implements "Time-Travel Debugging" - re-run agent

Report a penalty event to the trust backend.

Reset OTel state (for testing only).

Move a FAILED step back to PENDING for another execution attempt. Goes through ``transition()`` rather than mutating ``state`` direct

Resolve a dispute by replaying logs against Control Plane. Process: 1. Parse both flight recorder logs 2. Id

Resolve inheritance chains across SLO specs. Child specs override parent fields; unset fields inherit from parent. Returns new list with all

Retrieve episodes marked as failures.

Retrieve semantic rules from storage. Args: filters: Optional metadata filters limit: Maximum number

Retrieve episodes that are NOT failures.

Retrieve both successes and failures.

Call *fn* with exponential backoff + jitter on transient errors. Args: fn: Callable to invoke. max_retries: Number of retry attem

Rollback a rollout.

Rotate the value of an existing credential. Handle name is preserved so callers, prompts, and saved plans do not need to be updated.

Route request through Trust Gateway with security controls. Args: query: The user query to route use

Run trust verification for a pipeline stage. Args: component_name: Name of the component to verify. data: Data being

Simulates agent execution with tool use

Execute a sequence of (tool, params) through the kernel.

Run full audit pipeline

Run the CLI with the given arguments and return the exit code.

Send an order through the full governance pipeline.

Run static config checks plus live MCP tool inspection.

Returns True if the action is allowed.

Strip instruction tags from tool output and report stripped threats. Args: response_content: Raw tool output to sanitize.

Save a checkpoint record.

* Save approval request to disk.

Return obvious subprocess/process-spawning patterns in Python source.

Scan all tools from an MCP server. Args: server_name: Name of the MCP server. tools: List of tool dicts with keys: ``

Return a capability-deviation score in ``[0, 1]``. A value of 0 means no violations; 1 means every call was outside the declared prof

Create a scrubbed copy of the payload for logging. Redacts sensitive information.

Search tools by name, description, or tags. Args: query: Search query string include_all_versions: Whether to search

Send a JSON message over the WebSocket. Args: topic: Message topic. payload: Message data.

Serialize to bytes for authenticated encryption.

Set a session-level override for action classification.

Set the minimum score to pass (0-1).

Set up a test wisdom database with sample improvements. Args: temp_file: Path to temporary file for test database R

In a real system, this would be an LLM generating a plan. For this demo, we manually create plans to show the firewall in action.

Return a JSON-serializable snapshot of all metrics. Returns: Dictionary containing global and per-adapter metrics.

Capture a copy-on-write snapshot of the session's VM memory. Snapshots are kept in process memory and dropped when the session is des

Start a new child span. Args: operation_name: Name of the operation Returns: Ne

Suspend an agent (e.g., on anomaly detection).

Make a system call to the kernel.

Aggregate system health summary.

Serialize to a plain JSON-safe object (private key excluded).

Export this identity as a JWK (JSON Web Key, RFC 7517).

Return the ASGI callable.

Serialize to dictionary.

Convert this result to the legacy ``(allowed, reason)`` tuple.

Emit as a SLSA v1.0 / in-toto Statement provenance predicate.

Create an InputGuardrail that checks agent trust scores. Triggers tripwire if the agent's trust score is below the configured minimum.

Atomically update last_seen only if at least min_interval_seconds have elapsed since the last update. Returns True if updated, False i

Convert a CloudEvents type back to an AMB topic. Args: event_type: CloudEvents type (e.g., "dev.agent-os.fraud.alerts")

Update trust score based on a new execution. Args: success: Whether the execution was successful lat

Validate that the request complies with privacy policies. Returns: Tuple of (is_valid, error_message) is_val

Always passes.

Verify a peer agent's identity and return a trust result.

Verify an artifact's signature. Args: artifact_path: Path to the artifact file. signature: The Ed25519 signature byte

Verify the entire receipt hash chain + signatures.

Verify Ed25519-Timestamp auth header. Returns the agent DID. Format: Ed25519-Timestamp <did> <iso8601> <base64url(signature)> Spec: docs/spe

Verify multiple embedding pairs efficiently (CMVK-004). Processes all pairs with consistent settings and optional audit logging. Args: