Runtime Enforcement

Intent that only lives in a .intent file is advisory , a good intention a diff can quietly break. Intent that runs is load-bearing: it blocks the forbidden action when the software actually executes. compileGuard turns an intent into exactly that , a guard you drop into your application. Deterministic, no AI, and browser-safe (@skillstech/intentlang/core), so the same guard runs in a Node service or a browser app.

Two things an intent declares are deterministically enforceable at runtime, and the guard compiles both.

Secrets never leak

Every field the intent declares secret , a Secret / Password / Token type, a pii / sensitive data element, or a secret-looking name , the guard will mask. Wrap a logger or a response with redact and the secret cannot reach it:

import { compileGuard } from "@skillstech/intentlang/core";

const guard = compileGuard(intentSource);

logger.info("charging", guard.redact({ amount, paymentToken }));
// -> { amount: 50, paymentToken: "[redacted]" }   (deeply, through nested objects and arrays)

That is the never-rule "never expose the payment token in logs" enforced , not hoped.

Decisions become a hard gate

A decision is already executable; the guard makes it a production gate. assertAllowed runs the declared rules and throws if the intent denies the action:

try {
  guard.assertAllowed("RefundDecision", { amount, approved });
  await issueRefund();
} catch (e) {
  if (e.code === "INTENT_GUARD_DENIED") return reject(e.result); // the intent said no
  throw e;
}

The intent's rules , not a comment, not a code review , decide at runtime. By default the guard treats results like Deny, Blocked, Refuse, Reject, Escalate, NeedsReview as a block; pass denyResults to state the blocking results explicitly:

compileGuard(intentSource, { denyResults: ["Deny", "NeedsReview"] });

Preview what a guard enforces

intent guard <file> shows what the compiled guard would do , which fields it redacts and which decisions it gates , so it can be reviewed and audited before it ships:

intent guard Refund.intent
#   redacts fields   paymentToken, ssn
#   enforces decisions RefundDecision

Honest scope

The guard enforces what is structurally enforceable: the secrets the intent declares, and the decisions the intent makes executable. It does not turn a free-text never-rule into a runtime check on its own , that still needs the rule expressed as a decision or a typed field. What it gives you is the part that can be made load-bearing, made load-bearing , so the highest-cost mistakes (a leaked secret, a forbidden action) are blocked by the intent itself, in production.