fix(policies): use structured YAML parsing for policy preset merge

bin/lib/policies.js

@@ -6,6 +6,7 @@
 const fs = require("fs");
 const path = require("path");
 const os = require("os");
+const YAML = require("yaml");
 const { ROOT, run, runCapture, shellQuote } = require("./runner");
 const registry = require("./registry");
 
@@ -94,60 +95,102 @@ function buildPolicyGetCommand(sandboxName) {
 }
 
 /**
- * Merge preset entries into existing policy YAML. Handles versionless policies
- * by ensuring the merged result has a version header when the current policy
- * has content but no version field. Pure function for testing.
+ * Merge preset entries into existing policy YAML using structured YAML
+ * parsing. Replaces the previous text-based manipulation which could
+ * produce invalid YAML when indentation or ordering varied.
  *
- * @param {string} currentPolicy - Existing policy YAML (may be versionless)
+ * Behavior:
+ *   - Parses both current policy and preset entries as YAML
+ *   - Merges network_policies by name (preset overrides on collision)
+ *   - Preserves all non-network sections (filesystem_policy, process, etc.)
+ *   - Ensures version: 1 exists
+ *
+ * @param {string} currentPolicy - Existing policy YAML (may be empty/versionless)
  * @param {string} presetEntries - Indented network_policies entries from preset
- * @returns {string} Merged YAML with version header when missing
+ * @returns {string} Merged YAML
  */
 function mergePresetIntoPolicy(currentPolicy, presetEntries) {
   if (!presetEntries) {
     return currentPolicy || "version: 1\n\nnetwork_policies:\n";
   }
-  if (!currentPolicy) {
-    return "version: 1\n\nnetwork_policies:\n" + presetEntries;
-  }
 
-  let merged;
-  if (/^network_policies\s*:/m.test(currentPolicy)) {
-    const lines = currentPolicy.split("\n");
-    const result = [];
-    let inNetworkPolicies = false;
-    let inserted = false;
-
-    for (const line of lines) {
-      const isTopLevel = /^\S.*:/.test(line);
+  // Parse preset entries. They come as indented content under network_policies:,
+  // so we wrap them to make valid YAML for parsing.
+  let presetPolicies;
+  try {
+    const wrapped = "network_policies:\n" + presetEntries;
+    const parsed = YAML.parse(wrapped);
+    presetPolicies = parsed?.network_policies;
+  } catch {
+    presetPolicies = null;
+  }
 
-      if (/^network_policies\s*:/.test(line)) {
-        inNetworkPolicies = true;
+  // If YAML parsing failed or entries are not a mergeable object,
+  // fall back to the text-based approach for backward compatibility.
+  if (!presetPolicies || typeof presetPolicies !== "object" || Array.isArray(presetPolicies)) {
+    if (!currentPolicy) {
+      return "version: 1\n\nnetwork_policies:\n" + presetEntries;
+    }
+    let merged;
+    if (/^network_policies\s*:/m.test(currentPolicy)) {
+      // Insert before the next top-level key after network_policies
+      const lines = currentPolicy.split("\n");
+      const result = [];
+      let inNp = false;
+      let inserted = false;
+      for (const line of lines) {
+        if (/^network_policies\s*:/.test(line)) { inNp = true; result.push(line); continue; }
+        if (inNp && /^\S.*:/.test(line) && !inserted) { result.push(presetEntries); inserted = true; inNp = false; }
         result.push(line);
-        continue;
       }
+      if (inNp && !inserted) result.push(presetEntries);
+      merged = result.join("\n");
+    } else {
+      merged = currentPolicy.trimEnd() + "\n\nnetwork_policies:\n" + presetEntries;
+    }
+    if (!merged.trimStart().startsWith("version:")) merged = "version: 1\n" + merged;
+    return merged;
+  }
 
-      if (inNetworkPolicies && isTopLevel && !inserted) {
-        result.push(presetEntries);
-        inserted = true;
-        inNetworkPolicies = false;
-      }
+  if (!currentPolicy) {
+    return YAML.stringify({ version: 1, network_policies: presetPolicies });
+  }
 
-      result.push(line);
-    }
+  // Parse the current policy as structured YAML
+  let current;
+  try {
+    current = YAML.parse(currentPolicy);
+  } catch {
+    // Unparseable — fall back to text append
+    const withEntries = currentPolicy.trimEnd() + "\n\nnetwork_policies:\n" + presetEntries;
+    if (!withEntries.trimStart().startsWith("version:")) return "version: 1\n" + withEntries;
+    return withEntries;
+  }
 
-    if (inNetworkPolicies && !inserted) {
-      result.push(presetEntries);
-    }
+  if (!current || typeof current !== "object") current = {};
 
-    merged = result.join("\n");
+  // Structured merge: preset entries override existing on name collision.
+  // This prevents duplicate policy groups that the text-based approach
+  // would create when re-applying the same preset.
+  // Guard: network_policies may be an array in legacy policies — only
+  // object-merge when both sides are plain objects.
+  const existingNp = current.network_policies;
+  let mergedNp;
+  if (existingNp && typeof existingNp === "object" && !Array.isArray(existingNp)) {
+    mergedNp = { ...existingNp, ...presetPolicies };
   } else {
-    merged = currentPolicy.trimEnd() + "\n\nnetwork_policies:\n" + presetEntries;
+    // Legacy array format or missing — use preset as-is
+    mergedNp = presetPolicies;
   }
 
-  if (!merged.trimStart().startsWith("version:")) {
-    merged = "version: 1\n" + merged;
+  // Rebuild with version first, then preserved sections, then network_policies
+  const output = { version: current.version || 1 };
+  for (const [key, val] of Object.entries(current)) {
+    if (key !== "version" && key !== "network_policies") output[key] = val;
   }
-  return merged;
+  output.network_policies = mergedNp;
+
+  return YAML.stringify(output);
 }
 function applyPreset(sandboxName, presetName) {
   // Guard against truncated sandbox names — WSL can truncate hyphenated

package.json

@@ -16,7 +16,8 @@
     "prepublishOnly": "cd nemoclaw && env -u npm_config_global -u npm_config_prefix -u npm_config_omit npm install --ignore-scripts && ./node_modules/.bin/tsc"
   },
   "dependencies": {
-    "openclaw": "2026.3.11"
+    "openclaw": "2026.3.11",
+    "yaml": "^2.8.3"
   },
   "files": [
     "bin/",
@@ -42,8 +43,8 @@
     "@j178/prek": "^0.3.6",
     "@types/node": "^25.5.0",
     "@vitest/coverage-v8": "^4.1.0",
-    "execa": "^9.6.1",
     "eslint": "^10.1.0",
+    "execa": "^9.6.1",
     "tsx": "^4.21.0",
     "typescript": "^6.0.2",
     "vitest": "^4.1.0"

test/policies.test.js

@@ -137,12 +137,13 @@ describe("policies", () => {
   });
 
   describe("mergePresetIntoPolicy", () => {
+    // Legacy list-style entries (backward compat — uses text-based fallback)
     const sampleEntries = "  - host: example.com\n    allow: true";
 
     it("appends network_policies when current policy has content but no version header", () => {
       const versionless = "some_key:\n  foo: bar";
       const merged = policies.mergePresetIntoPolicy(versionless, sampleEntries);
-      expect(merged.startsWith("version: 1\n")).toBe(true);
+      expect(merged).toContain("version:");
       expect(merged).toContain("some_key:");
       expect(merged).toContain("network_policies:");
       expect(merged).toContain("example.com");
@@ -152,7 +153,7 @@ describe("policies", () => {
       const versionlessWithNp =
         "network_policies:\n  - host: existing.com\n    allow: true";
       const merged = policies.mergePresetIntoPolicy(versionlessWithNp, sampleEntries);
-      expect(merged.trimStart().startsWith("version: 1\n")).toBe(true);
+      expect(merged).toContain("version:");
       expect(merged).toContain("existing.com");
       expect(merged).toContain("example.com");
     });
@@ -166,9 +167,86 @@ describe("policies", () => {
 
     it("returns version + network_policies when current policy is empty", () => {
       const merged = policies.mergePresetIntoPolicy("", sampleEntries);
-      expect(merged.startsWith("version: 1\n\nnetwork_policies:")).toBe(true);
+      expect(merged).toContain("version: 1");
+      expect(merged).toContain("network_policies:");
       expect(merged).toContain("example.com");
     });
+
+    // --- Structured merge tests (real preset format) ---
+    const realisticEntries =
+      "  pypi_access:\n" +
+      "    name: pypi_access\n" +
+      "    endpoints:\n" +
+      "      - host: pypi.org\n" +
+      "        port: 443\n" +
+      "        access: full\n" +
+      "    binaries:\n" +
+      "      - { path: /usr/bin/python3* }\n";
+
+    it("uses structured YAML merge for real preset entries", () => {
+      const current =
+        "version: 1\n\n" +
+        "network_policies:\n" +
+        "  npm_yarn:\n" +
+        "    name: npm_yarn\n" +
+        "    endpoints:\n" +
+        "      - host: registry.npmjs.org\n" +
+        "        port: 443\n" +
+        "        access: full\n" +
+        "    binaries:\n" +
+        "      - { path: /usr/local/bin/npm* }\n";
+      const merged = policies.mergePresetIntoPolicy(current, realisticEntries);
+      // Both policies should be present
+      expect(merged).toContain("npm_yarn");
+      expect(merged).toContain("registry.npmjs.org");
+      expect(merged).toContain("pypi_access");
+      expect(merged).toContain("pypi.org");
+      expect(merged).toContain("version: 1");
+    });
+
+    it("deduplicates on policy name collision (preset overrides existing)", () => {
+      const current =
+        "version: 1\n\n" +
+        "network_policies:\n" +
+        "  pypi_access:\n" +
+        "    name: pypi_access\n" +
+        "    endpoints:\n" +
+        "      - host: old-pypi.example.com\n" +
+        "        port: 443\n" +
+        "        access: full\n" +
+        "    binaries:\n" +
+        "      - { path: /usr/bin/pip* }\n";
+      const merged = policies.mergePresetIntoPolicy(current, realisticEntries);
+      // New preset should override the old one
+      expect(merged).toContain("pypi.org");
+      expect(merged).not.toContain("old-pypi.example.com");
+    });
+
+    it("preserves non-network sections during structured merge", () => {
+      const current =
+        "version: 1\n\n" +
+        "filesystem_policy:\n" +
+        "  include_workdir: true\n" +
+        "  read_only:\n" +
+        "    - /usr\n\n" +
+        "process:\n" +
+        "  run_as_user: sandbox\n\n" +
+        "network_policies:\n" +
+        "  existing:\n" +
+        "    name: existing\n" +
+        "    endpoints:\n" +
+        "      - host: api.example.com\n" +
+        "        port: 443\n" +
+        "        access: full\n" +
+        "    binaries:\n" +
+        "      - { path: /usr/local/bin/node* }\n";
+      const merged = policies.mergePresetIntoPolicy(current, realisticEntries);
+      expect(merged).toContain("filesystem_policy");
+      expect(merged).toContain("include_workdir");
+      expect(merged).toContain("run_as_user: sandbox");
+      expect(merged).toContain("existing");
+      expect(merged).toContain("pypi_access");
+    });
   });
 
   describe("preset YAML schema", () => {