The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/gsc-cli script utilizes uvx to download and execute code directly from a remote Git repository (https://github.com/AminForou/mcp-gsc). Executing code from a non-trusted third-party repository at runtime allows for the execution of unverified logic.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch its core functionality from an external URL hosted on GitHub. While GitHub is a known platform, the specific repository is not associated with a trusted organization or the skill author.
[CREDENTIALS_UNSAFE]: The scripts/gsc-cli script contains a hardcoded absolute path to a sensitive service account credential file: /home/lamrin/paperclip/service_account_credentials.json. Referencing specific credential files in scripts can lead to unauthorized data exposure.
[COMMAND_EXECUTION]: The skill provides a shell wrapper that passes all user-supplied arguments ($@) directly to a subprocess via exec uvx mcp2cli. This pattern can be exploited if the underlying tool does not properly sanitize inputs.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting data from external Google Search Console API responses.
Ingestion points: Data retrieved from properties, search analytics, and URL inspection results (via SKILL.md and references/tool-map.md).
Boundary markers: None present; the instructions do not include delimiters or warnings to ignore embedded instructions in retrieved data.
Capability inventory: The agent has the ability to execute shell commands and modify Search Console state (if GSC_ALLOW_DESTRUCTIVE is enabled).
Sanitization: No sanitization or validation of the API-sourced content is performed before processing.

gsc-cli