The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where external content fetched via web_fetch (as described in technical.md and aeo.md) could contain hidden instructions that influence the agent's output or actions. 1. Ingestion points: Untrusted data enters the agent context through website crawling in technical.md and server log analysis in aeo.md. 2. Boundary markers: The instructions do not define explicit delimiters or 'ignore' warnings for the fetched content. 3. Capability inventory: The agent has access to bash, write_to_file, and replace_file_content (specified in agents/openai.yaml). 4. Sanitization: No sanitization or filtering of external content is mentioned.
[COMMAND_EXECUTION]: The skill instructs the agent to utilize the bash tool to run local Bun scripts for AEO and OpenGraph audits. While the scripts provided are local, the use of shell commands with arguments derived from user input or external web data presents a risk if not properly handled by the agent.
[EXTERNAL_DOWNLOADS]: The skill references external services like Keywords Everywhere and Open PageRank APIs, as well as the Common Crawl data archive, for SEO metrics and data enrichment.

seo