AI coding tool settings audit
Pick the kind of file, paste it, and get a checklist of the settings that weaken permissions, let commands or MCP servers run without review, or send prompts and code to telemetry. Every check uses only the setting names and meanings in the vendors' documentation, linked below.
Runs in your browser. Nothing you type or paste leaves this page.
What it checks
- Permission mode
- Claude Code sessions that start in bypassPermissions or auto mode, managed settings that don't set disableBypassPermissionsMode to "disable", and a skipped bypass confirmation. For Copilot, whether allow-all (bypass) mode is disabled and whether teams can override it.
- Allow and deny rules
- Rules that allow every shell command (Bash or Bash(*)), a shell or interpreter with any arguments, curl or wget, or any web address; deny lists with nothing for .env files, keys or credentials; extra directories that cover the whole disk or home directory.
- MCP servers
- enableAllProjectMcpServers, which approves every server a repository defines, and managed settings with no MCP server allowlist.
- Hooks
- Hooks that download and run scripts or make network requests, HTTP hooks without an allowedHttpHookUrls list, and, in managed settings, whether only your hooks may run.
- Telemetry
- Claude Code's OTEL_LOG_USER_PROMPTS, OTEL_LOG_ASSISTANT_RESPONSES, OTEL_LOG_TOOL_DETAILS, OTEL_LOG_TOOL_CONTENT and OTEL_LOG_RAW_API_BODIES; Copilot's capture classes, captureContent and lockCaptureContent; plain-HTTP collectors and credentials written into the file.
- Sandbox and policy
- Whether the sandbox is required and can be bypassed, Copilot's forceRemoteSettingsRefresh, remote control and computer use, and plugin marketplace allowlists.
- Keys
- Keys the documentation doesn't list are reported as not recognised, and keys that don't take effect from the file you picked (such as a managed-only key in project settings) are pointed out.
Settings combine: Claude Code merges user, project, local and managed files, and managed settings win. Audit the file people actually get, and remember a clean result only means none of these checks matched.
SpanDock's guardrails are rules you set once on the server, and SpanDock enforces them on each person's computer. For Claude Code, Codex and Gemini CLI it installs a hook that checks the prompt and every tool call before it runs, and puts the hook back if someone removes it. Guardrails are on the Enterprise plan.
Sources
- Claude Code: settings files and precedence (checked on 2026-10-22)
- Claude Code: all settings (settings reference) (checked on 2026-10-22)
- Claude Code: configure permissions (checked on 2026-10-22)
- Claude Code: hooks reference (checked on 2026-10-22)
- Claude Code: monitoring (OpenTelemetry) (checked on 2026-10-22)
- GitHub Copilot: enterprise managed settings reference (checked on 2026-10-22)