Struct Policy 

pub struct Policy {
    pub allow_effects: BTreeSet<String>,
    pub allow_fs_read: Vec<PathBuf>,
    pub allow_fs_write: Vec<PathBuf>,
    pub allow_net_host: Vec<String>,
    pub allow_proc: Vec<String>,
    pub allow_approval: Vec<String>,
    pub budget: Option<u64>,
}

Re-exported so embedders can build a handlers::State::policy_ceiling without taking a direct dependency on lex-runtime. Policy a program is run under. Empty allow_effects = pure-only execution.

Wildcard scopes (read before embedding): the scope lists (allow_fs_read, allow_fs_write, allow_net_host, allow_proc) follow an empty = allow ANY convention, not empty = deny. Granting a scoped effect in allow_effects while leaving its scope list empty therefore opens the unrestricted form (any path / host / binary). That’s intentional for trusted local use (lex run), but it’s a footgun for embedders that build a Policy from untrusted input. Such embedders should populate the scope list for every kind they grant, or call Policy::wildcard_scoped_grants to detect the wide-open ones and refuse them. (#552)

Fields

allow_effects: BTreeSet<String>

allow_fs_read: Vec<PathBuf>

Path scope for the [fs_read] effect. Empty = any path (wildcard), not deny — see the type-level note above (#552).

allow_fs_write: Vec<PathBuf>

Path scope for the [fs_write] effect. Empty = any path (wildcard), not deny — see the type-level note above (#552).

allow_net_host: Vec<String>

Per-host scope on the [net] effect. Empty = any host (when [net] is in allow_effects); non-empty = only requests to these hosts succeed. Hosts compare against the URL’s host substring (port-agnostic). Lets a tool be granted [net] but scoped to e.g. api.openai.com only — without this, [net] is a blank check to exfiltrate anywhere.

allow_proc: Vec<String>

Per-binary scope on the [proc] effect. Empty = ANY binary allowed once [proc] is granted (treat as a global escape hatch; only acceptable for trusted code). Non-empty = proc.spawn(cmd, args) must match cmd against the basename portion of one of these entries. Per-arg validation is the caller’s responsibility — see SECURITY.md’s “argument injection” note.

allow_approval: Vec<String>

Per-scope allowlist on the [approval] effect. Empty = any scope allowed once approval is granted (treat as a global human-escalation escape hatch; only acceptable for trusted code). Non-empty = approval.request(scope, reason) must match scope against one of these entries — lets an operator grant e.g. “payment approvals only” rather than a blanket human-in-the-loop channel.

budget: Option<u64>

Implementations

impl Policy

pub fn pure() -> Policy

pub fn wildcard_scoped_grants(&self) -> Vec<&'static str>

Report the granted scoped effects whose scope list is empty — i.e. the ones the runtime treats as unrestricted (“any”): proc (any binary), net (any host), fs_read / fs_write (any path). Returns an empty vec when no granted kind is left wide open.

Intended for embedders that expose execution to untrusted callers: build the effective Policy, then refuse to run (or loudly log) if this returns non-empty. Pure / time / rand grants never appear here — they have no scope. (#552)

pub fn permissive() -> Policy

Trait Implementations

impl Clone for Policy

fn clone(&self) -> Policy

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Policy

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

Formats the value using the given formatter.

impl Default for Policy

fn default() -> Policy

Returns the “default value” for a type.