The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill stores API keys in unencrypted JSON files (config.json, config.local.json). The configure.ps1 script prompts for these secrets and saves them in plain text, which is an unsafe practice for sensitive credentials.- [COMMAND_EXECUTION]: The documentation and installation instructions (in README.md, SKILL.md, and install.ps1) recommend running PowerShell with -ExecutionPolicy Bypass. This command disables local security restrictions that are meant to protect against the execution of untrusted scripts.- [COMMAND_EXECUTION]: The script scripts/grok_search.py disables SSL certificate verification when communicating with the API (verify=False and urllib3.disable_warnings). This bypasses critical security protocols and leaves the connection vulnerable to interception by attackers.- [DATA_EXFILTRATION]: The skill transmits search queries and user-provided API keys to an external endpoint over the network. While necessary for the skill's function, it involves sending sensitive data to a third-party server.- [PROMPT_INJECTION]: The skill definition contains aggressive behavioral instructions that mandate its use before the agent provides answers, which can be used to bypass default reasoning steps.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from search results.
Ingestion points: Search results are fetched from an external API in scripts/grok_search.py.
Boundary markers: No explicit delimiters or instructions are used to separate external content from the prompt.
Capability inventory: The skill has network access and file system access for configuration.
Sanitization: No validation or filtering is performed on data retrieved from external websites.

grok-search