@suveren/gateway 0.27.0 → 0.27.1

package/content/guides/6-risks.md

@@ -8,7 +8,7 @@ What can make a test say more than it should, and what to do about each. Read th
 | The working AI learns it is a test | It may behave differently | Never give it a `reporting` mandate during a run — its tools list the cases. Grant `reporting` only after the run, or to a separate AI session that only writes the report. Keep test words out of the brief |
 | Setup and work in the same session | Setup tools and the real replies stay in the working AI's context | Use separate conversations for setup and for work |
 | Mandates too loose | A wrong result can look like success | Start in review mode or with low caps; loosen by replacing the mandate |
-| Proposals are stored readable on the Authority Server | The content of every proposal, including intents, can be read there | Test data and renamed cases only — never real customer data in a setup |
+| The values a limit checks are readable on the Authority Server | Proposals are encrypted end to end since 0.27.0 — the server sees only the values it enforces limits on (e.g. an amount or a number of recipients) and fingerprints, never the content | Test data and renamed cases only in a setup; limits that refer to personal data are a profile question |
 | Setup limit too low | A clear without the following load leaves empty systems | Use a setup limit of 4–6 per day; clear and load in one go |
 | An AI that can control a browser or the screen | It could open the gateway's page, read the API key there and approve its own proposals — the person would no longer decide | Never open or operate the gateway's page, never handle the API key, never approve. Tell the person to approve themselves, and to keep the gateway page out of any browser an AI controls |
 | Calls around the Suveren tools | A terminal, HTTP or browser call to the gateway's ports is not governed by a mandate | Use only the Suveren tools. If something is missing, tell the person what — do not probe the gateway |

package/dist/mcp-server/http.mjs

@@ -4715,7 +4715,7 @@ function createGatedToolHandlerInner(tool, integrationManager2, state2) {
         try {
           const enrichedArgs = await attachImagePreview(args);
           const wireExecCtx = wireExecutionContext(auth.profileId, execution);
-          let recipientUserIds = [...auth.deferredCommitmentDomains];
+          let recipientUserIds = auth.deferredCommitmentDomains.filter((d) => d !== "owner");
           try {
             const summary = await state2.spClient.getAuthorizationSummary(authzId);
             if (summary?.created_by) recipientUserIds = [summary.created_by, ...recipientUserIds];
@@ -8162,7 +8162,7 @@ var DELEGATION_INTENT_HINT = [
   "",
   "Goal \u2014 Follow the setup guide: interview me about the company, the team and the work; prepare test data from real cases, renamed to .example names; propose the mandates and the working instructions (agent brief) that the AI doing the work will need.",
   "",
-  "Watch out \u2014 You only propose; nothing happens until I approve it in the Suveren Gateway. Ask, don't guess: never invent a rule, a limit or a step. One proposal at a time. Steps customers would see start with my review. Never use real names or customer data \u2014 proposals are stored readable on the Suveren server. Never give the working AI a reporting mandate during a test run. Keep the setup and the working AI in separate conversations.",
+  "Watch out \u2014 You only propose; nothing happens until I approve it in the Suveren Gateway. Ask, don't guess: never invent a rule, a limit or a step. One proposal at a time. Steps customers would see start with my review. Never use real names or customer data in a test run \u2014 test data only. Never give the working AI a reporting mandate during a test run. Keep the setup and the working AI in separate conversations.",
   "",
   // DL1 — outside simulation, the same mandate also covers ongoing, real-system
   // use: the Scope above (which profiles, which mode) is what actually bounds