If you discover a security vulnerability in the Google Ads Agent plugin, the MCP servers it ships, or the agents/skills/hooks it contains, please do not open a public issue. Instead, email the maintainer privately:
- Email:
security@ahmeego.com(orjohn@ahmeego.comif the security alias bounces) - PGP: available on request
You will receive an acknowledgement within 72 hours and a status update within 7 days.
We take seriously, in order of priority:
- Credential exfiltration — anything that exposes Google Ads OAuth tokens, refresh tokens, developer tokens, Cloudinary keys, or the Anthropic API key out of the local environment
- Unauthorized writes to Google Ads accounts — any path that lets the agent mutate spend without explicit user confirmation
- Prompt injection that bypasses safety hooks — the
PreToolUseGAQL validator and confirmation gates - MCP server vulnerabilities — RCE, SSRF, or path traversal in
googleadsagent-mcpor@googleadsagent/ad-creative-mcp - Supply-chain attacks — malicious dependencies introduced via the published PyPI/npm packages
- Issues in unrelated Google Ads scripts, the broader
agent-skillslibrary, or other projects in theitallstartedwithaideanamespace (file those there directly). - Issues that require an attacker to already have root on the user's machine or full access to their Google Ads MCC.
- Issues in Anthropic Claude Code itself — report those to security@anthropic.com.
We follow the Claude Code plugin security model:
- Hooks run as system commands and have full access to your shell. We ship only deterministic, side-effect-free hooks (
validate-gaql.jsandlog-tool-call.js). - MCP servers are launched via the package registries (
uvx,npx) using pinned major versions. - No hook references files outside
${CLAUDE_PLUGIN_ROOT}. - No secret is embedded in the plugin — all credentials come from environment variables.
- Pre-publish, every file is scanned for token-shaped strings (
sk-,ya29.,pypi-, etc.).
To audit what this plugin will execute on your machine:
# After install, plugins are cached here:
ls ~/.claude/plugins/cache/claude-community/googleadsagent
# Inspect hooks
cat ~/.claude/plugins/cache/claude-community/googleadsagent/hooks/hooks.json
# Inspect MCP server commands
cat ~/.claude/plugins/cache/claude-community/googleadsagent/.mcp.jsonWe follow a 90-day coordinated disclosure window from the date of acknowledgement, with extensions for complex issues. Reporters who follow this policy will be credited in the release notes (opt-in).