The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external websites, competitive content, and third-party platform results (e.g., G2, Reddit, LinkedIn). This creates a surface for indirect prompt injection where malicious instructions embedded in those untrusted external sources could attempt to influence the agent's behavior.
Ingestion points: Analysis of user-provided URLs and competitor pages to check for content extractability and citation presence (SKILL.md, references/platform-ranking-factors.md).
Boundary markers: There are no specific instructions for the agent to use delimiters or safety wrappers when processing external content to prevent command interpretation.
Capability inventory: The skill operates in an environment where the agent may have access to tools for web search, URL fetching, and terminal execution.
Sanitization: The skill lacks explicit instructions for sanitizing or escaping content fetched from the web before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill recommends using various third-party CLI tools and services that are not on the trusted vendor list, such as afdocs by Fern and monitoring platforms like Otterly AI, Peec AI, and ZipTie.
[COMMAND_EXECUTION]: The documentation provides multiple shell command templates (e.g., curl for checking robots headers, npx for site readiness audits). An agent with terminal access could attempt to execute these commands autonomously while performing a requested audit, leading to unverified code execution if the package sources are not properly vetted.

ai-seo