The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill targets and verifies the existence of sensitive credential files and environment variables. Evidence: Checks for 'GEMINI_API_KEY', 'GOOGLE_API_KEY', and the file '~/.gemini/oauth_creds.json' in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the package '@anthropic-ai/gemini-cli'. This is highly suspicious as Gemini is a Google product, while '@anthropic-ai' is the namespace for Anthropic. Recommending a package that mismatches a vendor with its product is a significant supply chain risk indicative of typosquatting or malicious intent.
[COMMAND_EXECUTION]: The skill executes multiple shell commands (which, where, gemini, claude) to check status. Crucially, the 'Safety and guardrails' section instructs the agent to run these commands 'outside sandbox' and with 'escalated permissions' to avoid permission errors, which is an instruction to bypass standard security boundaries.
[DATA_EXFILTRATION]: The workflow involves reading local credentials and subsequently performing network requests via the CLI tools. This pattern of credential access followed by network activity represents a risk for data exposure or exfiltration.
[REMEDIATION]: Restrict the skill to a sandboxed environment with no access to the user's home directory or sensitive environment variables. Verify the correct official package names for the Gemini CLI (e.g., via Google's official documentation) and avoid running any third-party CLI tools with escalated privileges.

check-collaborators