The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts such as scripts/dfs_client.py and scripts/report_publish.py using bash, passing user-provided competitor and user-domain names directly as arguments. This presents a potential command injection surface if these domain inputs are not properly sanitized before shell execution.- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch the content of the competitor's top-ranking pages to analyze content patterns. This untrusted external data is brought directly into the agent's context, creating an attack surface where malicious instructions hidden on a competitor's website could influence the agent's behavior.
Ingestion points: webfetch of competitor URLs identified in the ranking output.
Boundary markers: The skill does not specify any delimiters or safety instructions to the agent for handling the fetched content.
Capability inventory: The agent has the ability to execute local Python scripts via bash, perform web fetches, and read local configuration files like seo-project.yml.
Sanitization: There is no evidence of sanitization, filtering, or validation of the external web content before it is processed by the agent.

competitor-audit