oathe / scan / reports / caution
Is axtonliu/ai-pair safe?
Use with caution. axtonliu/ai-pair is not clearly unsafe, but the audit flagged risks to review before you install it. Oathe's behavioral security audit gave the ai-pair skill by axtonliu a trust score of 70/100 with 7 findings (1 critical, 1 high).
https://clawhub.ai/axtonliu/ai-pair
Is axtonliu/ai-pair safe to install?
This skill implements a multi-agent AI collaboration system that explicitly bypasses Claude Code's permission system to execute external CLI commands and access project files. While the core functionality appears legitimate for code review, the permission bypasses and external command execution create significant security risks.
What security issues were found in axtonliu/ai-pair?
Category Scores
Findings (7)
CRITICAL Explicit Permission Bypass -40 ▶
The skill explicitly instructs agents to use 'mode: bypassPermissions' to execute external CLI commands and access project files without normal permission checks.
Evidence: Launch 3 agents using the Agent tool with `subagent_type: "general-purpose"` and `mode: "bypassPermissions"` (required because reviewers need to execute external CLI commands and read project files).
MEDIUM External Command Execution Instructions -15 ▶
The skill instructs agents to execute external CLI commands (codex, gemini) which could be vectors for code execution if those tools are compromised.
Evidence: cat /tmp/review-input.txt | codex exec "Review this code..." and cat /tmp/review-content.txt | gemini -p "Review this content..."
MEDIUM Sensitive File Access -20 ▶
During installation, the skill accessed multiple sensitive files including .env, SSH keys, AWS credentials, and other authentication files.
Evidence: Monitoring detected access to /home/oc-exec/.env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, .config/gcloud/application_default_credentials.json
MEDIUM External CLI Command Execution -25 ▶
The skill's core functionality involves executing external CLI commands (codex, gemini) which represents a code execution risk if those tools are malicious or compromised.
Evidence: Agent prompt templates include instructions to execute 'codex exec' and 'gemini -p' commands with user-controlled input
HIGH Complex Agent Coordination Security Risk -30 ▶
The skill creates a complex multi-agent system with permission bypasses that could be used to orchestrate sophisticated attacks while appearing to perform legitimate code review functions.
Evidence: Coordinates 3 agents with bypassed permissions, file system access, and external command execution capabilities
LOW External Network Connections -5 ▶
The skill made connections to external services during installation, though these appear to be legitimate clawhub API calls.
Evidence: Network connections to 185.125.190.18:443, 185.125.188.54:443, 216.150.1.1:443 and curl to clawhub.ai API
LOW Normal Installation Pattern -10 ▶
Installation behavior was mostly normal with expected file creation and network activity for skill installation.
Evidence: Files created in designated directory, legitimate clawhub API interactions, no blocked connections
Should I install axtonliu/ai-pair?
Oathe's verdict for axtonliu/ai-pair is CAUTION with a trust score of 70/100. Recommendation: Install with caution.