The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external context files such as .agents/product-marketing.md, .claude/product-marketing.md, or product-marketing-context.md at runtime. These files serve as ingestion points for untrusted data that could contain malicious instructions (indirect prompt injection) intended to manipulate the agent's behavior. The skill instructions do not include explicit boundary markers or "ignore embedded instructions" warnings for these files. Additionally, the skill has the capability to interact with external tools such as semrush, ahrefs, and ga4, which could be exploited if an injected instruction directs the agent to perform unauthorized actions using these capabilities. No specific sanitization or validation of the ingested file content is mentioned in the skill definition.
[SAFE]: The skill references numerous external resources, including official platform documentation (OpenAI, Perplexity, Anthropic, Meta, Google), technological standards (AgentReady, llmstxt.org, Open Knowledge Format), and well-known industry tools (Vercel's is-agentic, Frase, Ahrefs, Semrush). These references target trusted organizations or well-known services and do not contribute to a negative verdict.
[SAFE]: The documentation includes several shell command examples such as npx is-agentic, npx afdocs, curl, and ssh. These are provided in markdown code blocks as educational instructions for the human user to perform manual diagnostic checks and are not configured for automated execution by the agent.

ai-seo