The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation includes commands for quick installation that pipe a remote script from GitHub directly into the bash interpreter (curl -fsSL ... | bash). It also includes instructions for downloading and running a PowerShell script from the same untrusted source. These scripts are hosted on a repository that is not identified as a trusted vendor or well-known service.
[PRIVILEGE_ESCALATION]: The provided installation instructions for Windows suggest using the Bypass execution policy for PowerShell (-ExecutionPolicy Bypass). This command allows the execution of scripts while ignoring the current security policy set on the machine, which can be used to run unverified and potentially harmful code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites as part of its SEO analysis workflow. This creates a surface for indirect prompt injection attacks if the audited content contains malicious instructions designed to manipulate the agent.
Ingestion points: External site content retrieved by scripts such as fetch_page.py and tools like read_url_content.
Boundary markers: The skill does not define specific delimiters or instructions to protect against instructions embedded in the fetched data during analysis.
Capability inventory: The skill can execute local Python scripts, perform network operations, and write multiple report files (JSON, HTML, Markdown) to the local filesystem.
Sanitization: No explicit sanitization or filtering of external content is documented before it is passed to the reasoning agent for audit assessment.

agentic-seo-skill