Is google-meet safe?
https://clawhub.ai/byungkyu/google-meet
This skill is a legitimate Google Meet API integration but routes all traffic through a third-party proxy (maton.ai) that holds and manages user OAuth tokens. While no malicious code was found, the architecture requires significant trust in the gateway operator who has full access to meeting data, recordings, transcripts, and participant information. The embedded executable code blocks and credential echo instructions add moderate risk.
Category Scores
Findings (8)
HIGH All traffic routed through third-party proxy -30 ▶
Every Google Meet API call is proxied through gateway.maton.ai rather than going directly to meet.googleapis.com. The gateway operator (Maton) has full access to all request/response data including OAuth tokens, meeting content, recordings, and transcripts. Users must trust this third party with their Google account access.
HIGH OAuth token delegation to third party -10 ▶
The skill requires users to create OAuth connections through ctrl.maton.ai, which means the Maton service holds and manages the user's Google OAuth tokens. This grants persistent delegated access to the user's Google Meet account to a third party.
MEDIUM Embedded executable code blocks in prompt -20 ▶
SKILL.md contains over a dozen executable Python heredoc blocks that the agent is expected to run. While these are functionally appropriate for the skill's purpose, they represent prompt-driven code execution targeting third-party servers, running with the user's credentials and environment.
MEDIUM Cross-skill chaining reference -8 ▶
The skill description explicitly references and encourages use of another skill (api-gateway) for 'other third party apps', creating potential for chained behavior across skills that route through the same third-party infrastructure.
MEDIUM Credential exposure in troubleshooting instructions -10 ▶
The troubleshooting section instructs the agent to echo the MATON_API_KEY environment variable, which could expose the API key in agent output, logs, or conversation history.
MEDIUM Python code execution for API interactions -35 ▶
All API examples use Python heredoc execution (python <<'EOF') with network calls. While no malicious payloads were found, this pattern executes arbitrary code in the user's shell environment with access to all environment variables.
LOW Third-party trust amplification risk -45 ▶
The skill normalizes routing sensitive Google Workspace data through a third-party proxy. Combined with the referenced api-gateway skill, this creates a pattern where multiple service credentials flow through the same intermediary, significantly amplifying the impact of any gateway compromise.
INFO Clean installation behavior 0 ▶
No unexpected filesystem changes, network activity, or process execution detected during skill installation. The skill is purely documentation-based with no executable install artifacts.