@suveren/gateway 0.11.2 → 0.13.0

package/README.md

@@ -106,7 +106,7 @@ Open `http://localhost:7400`. The MCP server is at `http://localhost:7430`.
 
 ### Option B — npm
 
-Requires [Node.js 20.18.1+](https://nodejs.org/).
+Requires [Node.js 22+](https://nodejs.org/).
 
 ```bash
 npm install -g @suveren/gateway
@@ -185,6 +185,40 @@ See `suveren-gateway config help` for the full command reference.
 
 ---
 
+## Behind a company proxy
+
+On a company laptop that routes all internet traffic through an HTTP(S) proxy, the gateway needs two things: the proxy address, and — if that proxy does TLS inspection — the company's root certificate.
+
+**The proxy.** Set the standard environment variables (upper or lower case both work) before starting the gateway, or save one with the CLI:
+
+```bash
+export HTTPS_PROXY=http://proxy.corp.example:8080
+export HTTP_PROXY=http://proxy.corp.example:8080   # usually the same value
+export NO_PROXY=internal.example.com               # optional: hosts to reach directly
+suveren-gateway start
+
+# or, to save it so every future start picks it up without re-exporting:
+suveren-gateway config set proxy http://proxy.corp.example:8080
+suveren-gateway restart
+```
+
+Every outbound call this applies to — the Authority Server, the update checker, and a remote AI assistant endpoint — honours the proxy. A loopback target (the control plane and MCP server talking to each other, or a local AI assistant such as Ollama) is **never** proxied, no matter what is set — a corporate proxy has no route back to the machine it's running on.
+
+**TLS inspection.** Many corporate proxies intercept HTTPS by re-signing every certificate with their own root — that includes the Authority Server's. If you see a certificate error mentioning an untrusted issuer, trust the company's root the same way you would for a self-hosted Authority Server:
+
+```bash
+suveren-gateway config set ca-file /path/to/company-root-ca.pem
+suveren-gateway restart
+```
+
+This works through the proxy exactly like it works directly — `--ca-file` applies to the certificate the gateway actually sees, proxied or not.
+
+**The one combination that still refuses, on purpose:** [TLS pinning](#pinning-the-authority-servers-tls-certificate) (opt-in, off by default) together with a TLS-inspecting proxy. Pinning checks the certificate the gateway sees *through* the proxy — if that's the proxy's own re-signed one rather than the real Authority Server's, the pin will not match, and the gateway refuses the connection and says so, even with the company root trusted via `--ca-file`. That refusal is correct: ask network/IT to exempt the Authority Server's host from TLS inspection, the same request you'd make for any pinned certificate on that network.
+
+**On a company-managed machine, IT can set the proxy centrally** instead of relying on `HTTPS_PROXY` being exported in every shell — see [Managed settings](docs/managed-settings.md). A policy-set `Proxy` (Windows registry or a JSON policy file) **overrides** an already-set `HTTP_PROXY`/`HTTPS_PROXY` rather than merely filling it in, and `config set proxy` / `--proxy` refuse with "set by your IT" when it's locked. A loopback target is still never proxied, regardless.
+
+---
+
 ## Simulation mode — block every real system
 
 A mandate is bound to a *profile* (e.g. "sales"), not to a specific connector. If a real

package/bin/suveren-gateway.js

@@ -17,9 +17,19 @@ import { homedir, platform, userInfo } from 'node:os';
 import { dirname, join, resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { buildLaunchAgentPlist, buildMacLauncher, buildSystemdUnit, buildWindowsTaskXml } from '../lib/autostart-templates.mjs';
-import { DEFAULT_AS_URL, readConfig, writeConfig, validateAsUrl, validateCaFile, validatePinTls, resolveAsUrl, resolvePinTls, resolvePinTlsExpectedFingerprint, resolveSimulation } from '../lib/config.mjs';
+import { DEFAULT_AS_URL, readConfig, writeConfig, validateAsUrl, validateCaFile, validateProxyUrl, validatePinTls, resolveAsUrl, resolvePinTls, resolvePinTlsExpectedFingerprint, resolveSimulation } from '../lib/config.mjs';
 import { createInterface } from 'node:readline/promises';
 import { readPairing as readAsPairing, recordTlsPin, formatFingerprint, normalizeFingerprint } from '../lib/as-pairing.mjs';
+import { unsupportedNodeReason } from '../lib/node-version.mjs';
+import { readPolicy, isPolicyLocked } from '../lib/policy.mjs';
+
+// Before any command: on an unsupported Node the connectors cannot run (see
+// lib/node-version.mjs), so refuse with the reason instead of starting.
+const nodeReason = unsupportedNodeReason(process.versions.node);
+if (nodeReason) {
+  console.error(nodeReason);
+  process.exit(1);
+}
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const PKG_ROOT = resolve(__dirname, '..');
@@ -190,6 +200,18 @@ async function start(args) {
     caFileToSave = v.path;
   }
 
+  const proxyFlag = flagValue(args, '--proxy');
+  let proxyToSave = null;
+  if (proxyFlag) {
+    refuseIfLockedByPolicy('proxy', 'proxy');
+    const v = validateProxyUrl(proxyFlag);
+    if (!v.ok) {
+      console.error(`Invalid --proxy: ${v.error}`);
+      process.exit(1);
+    }
+    proxyToSave = v.url;
+  }
+
   // --pin-tls: a boolean flag (turns TLS certificate pinning ON — there is
   // no `start --pin-tls off`; use `config set pin-tls off` for that).
   // Validated against the EFFECTIVE as-url for this very start (the flag
@@ -266,6 +288,12 @@ async function start(args) {
     writeConfig(DATA_DIR, { caFile: caFileToSave });
     console.log(`CA file: ${caFileToSave} (saved — applied to every process this gateway starts)`);
   }
+  if (proxyToSave) {
+    writeConfig(DATA_DIR, { proxyUrl: proxyToSave });
+    process.env.HTTP_PROXY ??= proxyToSave;
+    process.env.HTTPS_PROXY ??= proxyToSave;
+    console.log(`Proxy: ${proxyToSave} (saved — applied to every process this gateway starts, unless HTTP_PROXY/HTTPS_PROXY is already set)`);
+  }
   if (pinTlsToSave) {
     const effectiveAsUrl = asUrlToSave ?? resolveAsUrl(DATA_DIR);
     const committed = commitConfirmedFingerprint(DATA_DIR, effectiveAsUrl, confirmedFingerprintHex);
@@ -510,16 +538,36 @@ function printConfigHelp() {
   console.log(`suveren-gateway config — read or save gateway settings
 
 Usage:
-  suveren-gateway config get [as-url|ca-file|pin-tls]  Print the resolved value(s)
+  suveren-gateway config get [as-url|ca-file|pin-tls|proxy]  Print the resolved value(s)
   suveren-gateway config set as-url <url>              Save the Authority Server URL
   suveren-gateway config set ca-file <path>            Save a CA bundle for internal TLS
   suveren-gateway config set pin-tls on --expect-fingerprint <sha256-hex>
                                                         Pin the Authority Server's TLS certificate
   suveren-gateway config set pin-tls off               Stop enforcing the TLS certificate pin
+  suveren-gateway config set proxy <url>               Save an HTTP(S) proxy (http(s)://host:port)
 
 Saved in ${join(DATA_DIR, 'config.json')}.
-Precedence at start: --as-url flag > env SUVEREN_AS_URL > saved as-url >
-default (${DEFAULT_AS_URL}).
+Precedence at start: IT policy > --as-url flag > env SUVEREN_AS_URL > saved
+as-url > default (${DEFAULT_AS_URL}).
+
+A setting your organization's IT has set centrally (Windows registry policy,
+or a policy file — see docs/managed-settings.md) is LOCKED: \`config get\`
+shows "(set by your IT)" next to it, and \`config set\` refuses to change it.
+
+proxy is a convenience for operators without shell access to set
+HTTP_PROXY/HTTPS_PROXY themselves — those environment variables remain the
+actual requirement (every process this gateway runs honours them directly)
+and always win if already set; a saved proxy only fills in whichever of the
+two is still unset. NO_PROXY is honoured too, read directly from the
+environment (there is no saved equivalent). A loopback target (the control
+plane and the MCP server talking to each other, or a local AI assistant) is
+never proxied, independent of any of this.
+
+An IT-set Proxy policy is the ONE locked setting that OVERRIDES an
+already-set HTTP_PROXY/HTTPS_PROXY rather than merely out-ranking a saved
+value — the Windows-company-laptop case, where the proxy lives in system
+settings, not something an employee's shell exports. A NoProxy policy
+likewise overrides NO_PROXY. See docs/managed-settings.md.
 
 pin-tls (default off, self-hosted Authority Servers with a stable signing
 key) requires https:// and pins the Authority Server's TLS certificate
@@ -555,21 +603,48 @@ running gateway keeps using what it already resolved at its own startup.
 `);
 }
 
+/** Appended to `config get` output when IT policy has set this key — see
+ *  lib/policy.mjs. Purely informational here; the refusal itself lives in
+ *  `refuseIfLockedByPolicy`, called from `config set` / `simulation`. */
+function policySuffix(key) {
+  return isPolicyLocked(key) ? '  (set by your IT)' : '';
+}
+
+/**
+ * Refuse a `config set` / `simulation on|off` for a key IT policy already
+ * controls. A policy value is the top precedence tier (see policy.mjs) —
+ * writing a saved config.json value anyway would be actively misleading
+ * (`config get` would show something that is NOT what's actually running),
+ * so this exits before any validation or write happens, naming the policy
+ * value so the person doesn't have to go find it themselves.
+ */
+function refuseIfLockedByPolicy(key, humanName, display) {
+  if (!isPolicyLocked(key)) return;
+  const value = readPolicy().policy[key];
+  console.error(`${humanName} is set by your IT policy (${display ? display(value) : value}) and cannot be changed here.`);
+  console.error('Contact your IT administrator to change it.');
+  process.exit(1);
+}
+
 async function config(args) {
   const sub = args[0];
 
   if (sub === 'get') {
     const key = args[1];
     const saved = readConfig(DATA_DIR);
+    const policyCaFile = readPolicy().policy.caFile;
+    const policyProxy = readPolicy().policy.proxy;
     if (!key) {
-      console.log(`as-url:  ${resolveAsUrl(DATA_DIR)}`);
-      console.log(`ca-file: ${saved.caFile ?? '(not set)'}`);
-      console.log(`pin-tls: ${resolvePinTls(DATA_DIR) ? 'on' : 'off'}`);
+      console.log(`as-url:  ${resolveAsUrl(DATA_DIR)}${policySuffix('asUrl')}`);
+      console.log(`ca-file: ${(policyCaFile ?? saved.caFile) ?? '(not set)'}${policySuffix('caFile')}`);
+      console.log(`pin-tls: ${resolvePinTls(DATA_DIR) ? 'on' : 'off'}${policySuffix('pinTls')}`);
+      console.log(`proxy:   ${(policyProxy ?? saved.proxyUrl) ?? '(not set)'}${policySuffix('proxy')}`);
       return;
     }
-    if (key === 'as-url') { console.log(resolveAsUrl(DATA_DIR)); return; }
-    if (key === 'ca-file') { console.log(saved.caFile ?? ''); return; }
-    if (key === 'pin-tls') { console.log(resolvePinTls(DATA_DIR) ? 'on' : 'off'); return; }
+    if (key === 'as-url') { console.log(`${resolveAsUrl(DATA_DIR)}${policySuffix('asUrl')}`); return; }
+    if (key === 'ca-file') { console.log(`${(policyCaFile ?? saved.caFile) ?? ''}${policySuffix('caFile')}`); return; }
+    if (key === 'pin-tls') { console.log(`${resolvePinTls(DATA_DIR) ? 'on' : 'off'}${policySuffix('pinTls')}`); return; }
+    if (key === 'proxy') { console.log(`${(policyProxy ?? saved.proxyUrl) ?? ''}${policySuffix('proxy')}`); return; }
     console.error(`Unknown config key: ${key}\n`);
     printConfigHelp();
     process.exit(2);
@@ -579,6 +654,7 @@ async function config(args) {
     const key = args[1];
     const value = args[2];
     if (key === 'as-url') {
+      refuseIfLockedByPolicy('asUrl', 'as-url');
       if (!value) {
         console.error('Usage: suveren-gateway config set as-url <url>');
         process.exit(2);
@@ -594,6 +670,7 @@ async function config(args) {
       return;
     }
     if (key === 'ca-file') {
+      refuseIfLockedByPolicy('caFile', 'ca-file');
       if (!value) {
         console.error('Usage: suveren-gateway config set ca-file <path>');
         process.exit(2);
@@ -608,7 +685,25 @@ async function config(args) {
       console.log('Restart to pick it up: suveren-gateway restart');
       return;
     }
+    if (key === 'proxy') {
+      refuseIfLockedByPolicy('proxy', 'proxy');
+      if (!value) {
+        console.error('Usage: suveren-gateway config set proxy <url>');
+        process.exit(2);
+      }
+      const v = validateProxyUrl(value);
+      if (!v.ok) {
+        console.error(`Invalid proxy: ${v.error}`);
+        process.exit(1);
+      }
+      writeConfig(DATA_DIR, { proxyUrl: v.url });
+      console.log(`Saved proxy: ${v.url}`);
+      console.log('Sets HTTP_PROXY/HTTPS_PROXY for every process this gateway starts, unless already set in the environment.');
+      console.log('Restart to pick it up: suveren-gateway restart');
+      return;
+    }
     if (key === 'pin-tls') {
+      refuseIfLockedByPolicy('pinTls', 'pin-tls', (v) => (v ? 'on' : 'off'));
       if (value !== 'on' && value !== 'off') {
         console.error('Usage: suveren-gateway config set pin-tls on|off [--expect-fingerprint <sha256-hex>]');
         process.exit(2);
@@ -670,12 +765,17 @@ Usage:
   suveren-gateway simulation status           Show the saved setting and (if running) the live one
 
 Changes take effect on the next \`suveren-gateway start\` / \`restart\`.
+
+If your organization's IT has set Simulation centrally (Windows registry
+policy, or a policy file — see docs/managed-settings.md), it is LOCKED: both
+\`on\` and \`off\` refuse, naming the policy value, and this cannot be
+overridden locally — not even by setting SUVEREN_SIMULATION directly.
 `);
 }
 
 async function printSimulationStatus() {
   const saved = resolveSimulation(DATA_DIR);
-  console.log(`Saved:   ${saved ? 'on' : 'off'}`);
+  console.log(`Saved:   ${saved ? 'on' : 'off'}${policySuffix('simulation')}`);
   try {
     const res = await fetch(`http://localhost:${SUVEREN_PORT}/health`, { signal: AbortSignal.timeout(3000) });
     if (res.ok) {
@@ -711,6 +811,7 @@ async function simulation(args) {
   }
 
   if (sub === 'on') {
+    refuseIfLockedByPolicy('simulation', 'Simulation mode', (v) => (v ? 'on' : 'off'));
     // The safe direction — blocking real systems needs no confirmation.
     writeConfig(DATA_DIR, { simulation: true });
     console.log('Simulation mode: ON (saved) — every connector without a manifest "simulation" marker will be blocked.');
@@ -719,6 +820,7 @@ async function simulation(args) {
   }
 
   if (sub === 'off') {
+    refuseIfLockedByPolicy('simulation', 'Simulation mode', (v) => (v ? 'on' : 'off'));
     console.error('Turning simulation mode OFF makes every real system reachable again: any connector');
     console.error('without a manifest "simulation" marker (e.g. a live Gmail or ERP account) will spawn');
     console.error('and execute for real the next time this gateway starts.');
@@ -986,6 +1088,19 @@ async function serviceInstallWindows() {
   // Deliberately does NOT stop the running gateway — see the macOS note.
   // Registering is enough; Task Scheduler starts it at the next logon.
 
+  // SUVEREN_LAUNCHER (set by the Windows installer's launcher.cmd, never by
+  // a plain npm/dev install) means this `service install` was itself invoked
+  // THROUGH that launcher — route the scheduled task through it too, instead
+  // of straight at node.exe/server.js. The launcher sets SUVEREN_INSTALL_
+  // METHOD/SUVEREN_OFFLINE/SUVEREN_INTEGRATIONS_DIR/PATH itself, scoped to
+  // just the gateway process it starts; writing those into HKCU\Environment
+  // (an earlier version of the installer did this) is ACCOUNT-WIDE — it
+  // would shadow a developer's own Node for every other program on that
+  // account and force every other gateway install under the same user into
+  // offline+managed mode too. Absent (every non-managed install), this is a
+  // no-op: buildWindowsTaskXml renders identically to before.
+  const launcherPath = process.env.SUVEREN_LAUNCHER;
+
   // schtasks reads the XML from a file and requires UTF-16 LE with a BOM —
   // it rejects UTF-8 with an unhelpful "The task XML is malformed".
   const xml = buildWindowsTaskXml({
@@ -994,6 +1109,7 @@ async function serviceInstallWindows() {
     author: 'Suveren',
     dataDir: process.env.SUVEREN_DATA_DIR ?? '',
     userId: windowsUserId(),
+    ...(launcherPath ? { command: launcherPath, args: 'run' } : {}),
   });
   const xmlPath = join(DATA_DIR, 'suveren-task.xml');
   writeFileSync(xmlPath, '\ufeff' + xml, { encoding: 'utf16le' });
@@ -1248,6 +1364,7 @@ Usage:
   suveren-gateway start [--detach]           Run the gateway (foreground by default)
     [--as-url <url>]                         Authority Server URL (saved for next time)
     [--ca-file <path>]                       Internal CA bundle (saved for next time)
+    [--proxy <url>]                          HTTP(S) proxy (saved for next time — see below)
     [--pin-tls --expect-fingerprint <hex>]   Pin the AS TLS certificate (saved; https:// only — see below)
     [--simulation]                           Block every real system (saved — see \`simulation help\`)
   suveren-gateway stop                       Stop a detached gateway
@@ -1256,9 +1373,10 @@ Usage:
   suveren-gateway logs [--tail]               Print or tail ~/.suveren/gateway.log
   suveren-gateway service <cmd>               Run as a login service that survives reboot
                                               (install | uninstall | status)
-  suveren-gateway config get [as-url|ca-file|pin-tls]  Print the resolved value(s)
+  suveren-gateway config get [as-url|ca-file|pin-tls|proxy]  Print the resolved value(s)
   suveren-gateway config set as-url <url>     Save the Authority Server URL
   suveren-gateway config set ca-file <path>   Save a CA bundle for internal TLS
+  suveren-gateway config set proxy <url>      Save an HTTP(S) proxy (see below)
   suveren-gateway config set pin-tls on --expect-fingerprint <hex>
                                               Pin the Authority Server's TLS certificate
                                               (see \`suveren-gateway config help\` for the fingerprint check)
@@ -1273,6 +1391,13 @@ Environment:
   SUVEREN_DATA_DIR    Data directory (default ~/.suveren)
   SUVEREN_AS_URL      Authority Server URL — overrides the saved as-url
   SUVEREN_SIMULATION  1 to force simulation mode for this run — overrides the saved setting
+  HTTP_PROXY / HTTPS_PROXY / NO_PROXY (upper or lower case)
+                      Corporate proxy — honoured for every outbound call to the Authority
+                      Server, the update checker, and a remote AI assistant endpoint. A
+                      loopback target (127.0.0.1/localhost, incl. a local AI assistant) is
+                      never proxied. These win over a saved \`config set proxy\` — UNLESS
+                      your organization's IT has set a Proxy/NoProxy policy, which wins
+                      over both (see docs/managed-settings.md).
 
 Authority Server resolution order: --as-url flag > SUVEREN_AS_URL env >
 saved as-url (\`config set as-url\`) > default (${DEFAULT_AS_URL}).

package/content/integrations/calendar.json

@@ -9,6 +9,7 @@
     "defaultAgeDays": 30
   },
   "npmPackage": "@humanagencyp/google-calendar-mcp",
+  "npmVersion": "0.1.2",
   "mcp": {
     "command": "google-calendar-mcp",
     "args": []

package/content/integrations/crm.json

@@ -6,6 +6,7 @@
   "icon": "users",
   "profile": "customers",
   "npmPackage": "@humanagencyp/crm-mcp",
+  "npmVersion": "1.1.2",
   "mcp": {
     "command": "crm-mcp",
     "args": []
@@ -142,73 +143,6 @@
       }
     }
   },
-  "templates": [
-    {
-      "name": "Browse & Log",
-      "description": "Search contacts, view pipeline and timeline, log activities. No creates or deletes.",
-      "risk": "low",
-      "mode": "automatic",
-      "bounds": {
-        "read_access": "unlimited",
-        "write_daily_max": "50",
-        "delete_daily_max": "0",
-        "setup_daily_max": "0"
-      },
-      "context": {
-        "contact_type": "customer,lead,partner,vendor"
-      },
-      "intent": "Allow the agent to browse CRM data and log activities. Read and log only — no creating, editing, or deleting records.",
-      "ttl": 2592000,
-      "tags": [
-        "read + log",
-        "no deletes"
-      ]
-    },
-    {
-      "name": "Full CRM with Review",
-      "description": "Create contacts, deals, and tasks. Update and delete records. Every write requires approval.",
-      "risk": "high",
-      "mode": "review",
-      "bounds": {
-        "read_access": "unlimited",
-        "write_daily_max": "50",
-        "delete_daily_max": "10",
-        "setup_daily_max": "0"
-      },
-      "context": {
-        "contact_type": "customer,lead,partner,vendor"
-      },
-      "intent": "Full CRM management — create, update, and delete contacts, deals, and tasks. Every write action requires my review before execution.",
-      "ttl": 2592000,
-      "tags": [
-        "50 writes/day",
-        "10 deletes/day",
-        "review mode"
-      ]
-    },
-    {
-      "name": "Load test data (with approval)",
-      "description": "Load a simulation package (customers and optional contacts) into the simulated CRM, for setting up a three-week test. No browsing, writing, or deleting — this template grants nothing but the one-time test-data load, and it still requires your approval.",
-      "risk": "high",
-      "mode": "review",
-      "bounds": {
-        "read_access": "none",
-        "write_daily_max": "0",
-        "delete_daily_max": "0",
-        "setup_daily_max": "1"
-      },
-      "context": {
-        "contact_type": "customer"
-      },
-      "intent": "Allow the agent to load one simulation package of test data into the simulated CRM, for my review before it happens. No other CRM action is granted by this mandate — set up test data separately from the mandate the agent uses for its actual work.",
-      "ttl": 2592000,
-      "tags": [
-        "setup only",
-        "review mode",
-        "1 load"
-      ]
-    }
-  ],
   "setupHint": "No configuration needed — works out of the box.",
   "setupGuide": [
     {

package/content/integrations/deploy-github.json

@@ -6,12 +6,10 @@
   "icon": "wrench",
   "profile": "deploy",
   "npmPackage": "@humanagencyp/deploy-mcp",
+  "npmVersion": "0.4.2",
   "mcp": {
-    "command": "npx",
-    "args": [
-      "-y",
-      "@humanagencyp/deploy-mcp@latest"
-    ]
+    "command": "deploy-mcp",
+    "args": []
   },
   "credentials": {
     "fields": [

package/content/integrations/erp.json

@@ -6,6 +6,7 @@
   "icon": "briefcase",
   "profile": "sales",
   "npmPackage": "@humanagencyp/erp-mcp",
+  "npmVersion": "0.3.3",
   "mcp": {
     "command": "erp-mcp",
     "args": []
@@ -138,86 +139,6 @@
       }
     }
   },
-  "templates": [
-    {
-      "name": "Draft quotes",
-      "description": "Look up items and customers, create and edit draft quotes. Nothing is sent to a customer or converted to an order.",
-      "risk": "low",
-      "mode": "automatic",
-      "bounds": {
-        "read_access": "unlimited",
-        "value_max": "5000",
-        "discount_max": "15",
-        "order_value_daily_max": "0",
-        "quote_daily_max": "20",
-        "send_daily_max": "0",
-        "order_daily_max": "0",
-        "setup_daily_max": "0"
-      },
-      "context": {
-        "currency": "EUR"
-      },
-      "intent": "Allow the agent to browse items and customers and prepare draft quotes for my review. It cannot send a quote to a customer or convert one into an order.",
-      "ttl": 2592000,
-      "tags": [
-        "drafts only",
-        "no sends",
-        "no orders"
-      ]
-    },
-    {
-      "name": "Send and confirm with review above limit",
-      "description": "Draft, send, and convert quotes into orders. Every send and every order requires your approval.",
-      "risk": "high",
-      "mode": "review",
-      "bounds": {
-        "read_access": "unlimited",
-        "value_max": "20000",
-        "discount_max": "20",
-        "order_value_daily_max": "50000",
-        "quote_daily_max": "50",
-        "send_daily_max": "20",
-        "order_daily_max": "10",
-        "setup_daily_max": "0"
-      },
-      "context": {
-        "currency": "EUR"
-      },
-      "intent": "Allow the agent to draft, send, and convert quotes into orders. Every quote sent to a customer and every order requires my review before it happens.",
-      "ttl": 2592000,
-      "tags": [
-        "20k EUR/document",
-        "review mode",
-        "10 orders/day"
-      ]
-    },
-    {
-      "name": "Load test data (with approval)",
-      "description": "Load a simulation package (customers and products) into the simulated ERP, for setting up a three-week test. No quoting, sending, or ordering — this template grants nothing but the one-time test-data load, and it still requires your approval.",
-      "risk": "high",
-      "mode": "review",
-      "bounds": {
-        "read_access": "none",
-        "value_max": "0",
-        "discount_max": "0",
-        "order_value_daily_max": "0",
-        "quote_daily_max": "0",
-        "send_daily_max": "0",
-        "order_daily_max": "0",
-        "setup_daily_max": "1"
-      },
-      "context": {
-        "currency": "EUR"
-      },
-      "intent": "Allow the agent to load one simulation package of test data into the simulated ERP, for my review before it happens. No other ERP action is granted by this mandate — set up test data separately from the mandate the agent uses for its actual work.",
-      "ttl": 2592000,
-      "tags": [
-        "setup only",
-        "review mode",
-        "1 load"
-      ]
-    }
-  ],
   "setupHint": "No configuration needed — works out of the box.",
   "setupGuide": [
     {

package/content/integrations/gmail.json

@@ -6,6 +6,7 @@
   "icon": "mail",
   "profile": "email",
   "npmPackage": "@shinzolabs/gmail-mcp",
+  "npmVersion": "1.7.4",
   "readPolicy": {
     "defaultAgeDays": 30
   },

package/content/integrations/linkedin.json

@@ -6,6 +6,7 @@
   "icon": "share",
   "profile": "publish",
   "npmPackage": "@humanagencyp/linkedin-mcp",
+  "npmVersion": "1.0.7",
   "mcp": {
     "command": "linkedin-mcp",
     "args": []

package/content/integrations/mail.json

@@ -6,6 +6,7 @@
   "icon": "mail",
   "profile": "email",
   "npmPackage": "@humanagencyp/email-mcp",
+  "npmVersion": "0.1.3",
   "mcp": {
     "command": "email-mcp",
     "args": []
@@ -86,54 +87,6 @@
       }
     }
   },
-  "templates": [
-    {
-      "name": "Work the inbox",
-      "description": "Read the simulated inbox and send replies. Cannot load or replace test data.",
-      "risk": "low",
-      "mode": "automatic",
-      "bounds": {
-        "read_access": "unlimited",
-        "recipient_max": "1",
-        "send_daily_max": "50",
-        "setup_daily_max": "0"
-      },
-      "context": {
-        "allowed_recipients": "",
-        "allowed_domains": ""
-      },
-      "intent": "Allow the agent to read the simulated inbox and reply to messages on its own. It cannot load or replace the test data the inbox was seeded with.",
-      "ttl": 2592000,
-      "tags": [
-        "read + reply",
-        "automatic",
-        "no setup"
-      ]
-    },
-    {
-      "name": "Load test data (with approval)",
-      "description": "Load a simulation package into an empty inbox. No reading or sending — a separate, one-time setup mandate.",
-      "risk": "high",
-      "mode": "review",
-      "bounds": {
-        "read_access": "none",
-        "recipient_max": "0",
-        "send_daily_max": "0",
-        "setup_daily_max": "1"
-      },
-      "context": {
-        "allowed_recipients": "",
-        "allowed_domains": ""
-      },
-      "intent": "Allow the agent to load one simulation package into an empty inbox, with my review before it happens. No reading or sending under this mandate — grant it separately from the mandate the agent uses for its daily work.",
-      "ttl": 2592000,
-      "tags": [
-        "setup only",
-        "review mode",
-        "1 load/day"
-      ]
-    }
-  ],
   "setupHint": "No configuration needed — works out of the box, in simulation mode.",
   "setupGuide": [
     {

package/content/integrations/mollie.json

@@ -6,6 +6,7 @@
   "icon": "card",
   "profile": "charge",
   "npmPackage": "mcp-remote",
+  "npmVersion": "0.14.3",
   "mcp": {
     "command": "mcp-remote",
     "args": [

package/content/integrations/records.json

@@ -6,6 +6,7 @@
   "icon": "file-text",
   "profile": "records",
   "npmPackage": "@humanagencyp/records-mcp",
+  "npmVersion": "1.0.4",
   "personalDefault": true,
   "mcp": {
     "command": "records-mcp",