The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation instructions require cloning a repository from https://github.com/SeoToolkit/web-analytics-agent-skill.git. This repository is not owned by the stated author (Aradotso) and is not recognized as a trusted organization or well-known service.
[REMOTE_CODE_EXECUTION]: The skill requires the user to execute downloaded scripts, including auth_google.py, analyze_gsc.py, and ga4_both.py. Notably, it includes a shell script (run_all.sh) described as automatically checking and installing dependencies, which facilitates the execution of unverified remote code.
[COMMAND_EXECUTION]: The primary functionality of the skill relies on executing various local Python scripts and a shell script to manage authentication and data retrieval, granting the downloaded code extensive control over the local execution environment.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by fetching untrusted search query data and traffic metadata from Google and Bing APIs and outputting it into reports for agent processing.
Ingestion points: Search console queries, GA4 session sources, and Bing query statistics are ingested via the scripts and Python API examples.
Boundary markers: No delimiters or instructions are used to distinguish untrusted external data from system instructions.
Capability inventory: The skill has capabilities for file system writes (reports/ directory), network access (requests), and shell script execution (run_all.sh).
Sanitization: There is no evidence of filtering or escaping for external data fields such as query strings or source names.

web-analytics-agent-skill