The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the agent's shell to execute Python scripts (dfs_client.py, report_publish.py) using arguments derived directly from user input.
Evidence: python scripts/dfs_client.py ranked --target "<domain>" --limit 200
Evidence: python scripts/report_publish.py <report>.md where the filename is constructed using the <domain> input.
The construction of these shell commands lacks explicit sanitization or validation instructions for the <domain> and <suspect-page-url> inputs, which could lead to command injection if malicious shell metacharacters are provided.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and analyze content from external, untrusted websites, creating a surface for indirect prompt injection.
Ingestion points: The skill uses the webfetch tool to retrieve content from external pages for audit purposes (e.g., checking for bylines, branding, and content quality).
Boundary markers: The instructions do not define delimiters or provide "ignore embedded instructions" warnings for the agent when processing this retrieved content.
Capability inventory: The skill has access to shell execution via the included Python scripts in the scripts/ directory.
Sanitization: There are no instructions to sanitize or validate the content fetched from external domains before the agent processes it or includes it in the generated reports.

parasite-seo-check