Struct GovernanceEngine 

pub struct GovernanceEngine { /* private fields */ }

Authorization engine backed by Cedar policies.

Thread-safe — all methods take &self and the policy set is behind a [RwLock] to allow runtime reload without downtime.

Example

let engine = GovernanceEngine::load_from_dir("policies").await?;
let allowed = engine.is_tool_allowed("agent-1", "web_search").await;
assert!(allowed);

Implementations

impl GovernanceEngine

pub fn new() -> Self

Create a new governance engine with an empty (permissive) policy set.

With no policies loaded, all requests are implicitly denied by Cedar. Use Self::with_default_permit for a permissive default.

pub fn with_default_permit() -> Result<Self>

Create a governance engine with a default “permit all” policy.

This is the recommended starting point — agents can execute any tool until restrictive policies are explicitly added.

pub async fn load_from_dir(dir: impl AsRef<Path>) -> Result<Self>

Load policies from a directory of .cedar files.

All *.cedar files in the directory are concatenated and parsed as a single policy set.

pub async fn reload(&self) -> Result<()>

Reload policies from the configured directory.

This is a hot-reload: the new policies take effect immediately for all subsequent authorization checks, without restarting the server.

pub async fn is_tool_allowed(&self, agent_id: &str, tool_name: &str) -> bool

Check whether a tool execution is allowed.

Shorthand for building a tool execution request and evaluating it.

pub async fn tool_decision( &self, agent_id: &str, tool_name: &str, risk_requires_approval: bool, ) -> ToolGovernanceDecision

Evaluate a tool call into the complete runtime governance outcome.

pub async fn is_allowed( &self, agent_id: &str, action: &str, resource: &str, ) -> bool

Check whether a generic action is allowed.

pub async fn is_skill_mutation_allowed( &self, agent_id: &str, action: &str, skill_id: &str, context_json: &str, ) -> bool

Check whether a skill mutation is allowed.

Skill mutations carry rich context (environment, confidence scores, validation status) that Cedar policies evaluate against. Used by the self-learning pipeline to gate prompt optimization, skill generation, and skill promotion.

Arguments

• agent_id — The agent/optimizer attempting the mutation.
• action — One of: skill.mutate, skill.generate, skill.promote, trace.capture.
• skill_id — The skill being operated on.
• context_json — JSON object with mutation context. Policies evaluate fields like environment, confidence_score, validation_passed, human_approved, test_pass_rate, governance_consent.

Example

let allowed = engine.is_skill_mutation_allowed(
    "pmpo-optimizer",
    "skill.mutate",
    "iterative-evolver",
    r#"{"environment": "staging", "validation_passed": true}"#,
).await;

pub async fn evaluate_full(&self, request: &Request) -> Response

Evaluate a Cedar request against the current policy set.

Returns the full [Response] for detailed diagnostics.

pub async fn evaluate(&self, request: &Request) -> Decision

Evaluate a Cedar request, returning only the decision.

pub async fn policy_count(&self) -> usize

Get the number of currently loaded policies.

Trait Implementations

impl Debug for GovernanceEngine

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for GovernanceEngine

fn default() -> Self

Returns the “default value” for a type.