@northbridge-security/secureai 0.1.13

package/.claude/README.md

@@ -0,0 +1,122 @@
+# Claude Configuration
+
+The global configuration ensures, and cannot be overridden, that:
+
+- Harmful or dangerous commands are blocked
+- PII and secrets are not read or shared
+- Claude does not add co-authoring information in commits
+
+Standard setup:
+
+```json
+{
+  "hooks": {
+    "PreToolUse": [
+      {
+        "hooks": [
+          {
+            "command": "secureai guard --stdin --agent claude",
+            "type": "command"
+          }
+        ],
+        "matcher": "*"
+      }
+    ],
+    "UserPromptSubmit": [
+      {
+        "hooks": [
+          {
+            "command": "secureai prompt --stdin --agent claude",
+            "type": "command"
+          }
+        ]
+      }
+    ]
+  },
+  "includeCoAuthoredBy": false,
+  "permissions": {
+    "deny": [
+      "Bash(npm publish:*)",
+      "Bash(git push:*)",
+      "Bash(npx publish:*)",
+      "Bash(GH_TOKEN=:*)",
+      "Bash(export GH_TOKEN:*)",
+      "Bash(gh:*)",
+      "Edit(.env)",
+      "Edit(.env.*)",
+      "Write(.env)",
+      "Write(.env.*)",
+      "Bash(cat ~/.claude:*)",
+      "Bash(cat /Users/*/.claude:*)",
+      "Bash(cat $HOME/.claude:*)",
+      "Read(~/.claude/.credentials.json)",
+      "Read(~/.claude/credentials.json)",
+      "Read(~/.config/claude/*)",
+      "Bash(claude mcp get:*)"
+    ],
+    "disableBypassPermissionsMode": "disable"
+  }
+}
+```
+
+The local configuration focuses on productivity and allows uninterrupted use for agents of:
+
+- Use all Secure AI MCP features expect branch merging, which requires approval
+- Uses permission lockdown as mode `Strict` to only allow our MCP server to be used
+- Uses `allow` patterns to avoid repeated permission approvals
+
+This setup gives a balance between strict security and productivity.
+
+Standard setup:
+
+```json
+{
+  "permissions": {
+    "allow": ["Read", "Edit", "Write", "Bash", "WebSearch", "WebFetch", "mcp__secureai-mcp__*"],
+    "ask": [
+      "Edit(.pii)",
+      "Write(.pii)",
+      "Edit(.claude/*)",
+      "Edit(.cursor/*)",
+      "Edit(.husky/*)",
+      "Edit(.vscode/*)",
+      "mcp__secureai-mcp__plan_merge"
+    ],
+    "deny": ["Read(.env*)", "Edit(.env*)", "Write(.env*)"]
+  }
+}
+```
+
+- Allow reading, editing and creating files except settings and hooks files that require approval
+- Allow searching and fetching web content
+- Do not read or edit .env secret files
+
+Strict setup:
+
+```json
+{
+  "enableAllProjectMcpServers": false,
+  "enabledMcpjsonServers": ["secureai-mcp"],
+  "permissions": {
+    "allow": [
+      "Read",
+      "Edit(.worktree/**)",
+      "Edit(.tmp/**)",
+      "Write(.worktree/**)",
+      "Write(.tmp/**)",
+      "Bash",
+      "WebSearch",
+      "WebFetch",
+      "mcp__secureai-mcp__*"
+    ],
+    "ask": ["Edit", "Write", "mcp__secureai-mcp__plan_merge"],
+    "deny": ["Read(.env*)", "Edit(.env*)", "Write(.env*)"]
+  }
+}
+```
+
+- Allow edits unrestricted inside the .tmp and .worktree folders, ask for edits on all other locations
+- Disable MCP servers outside of Secure AI
+- Allow searching and fetching web content
+- Do not read or edit .env secret files
+