The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from external websites or user-provided HTML files. An attacker who controls content on the crawled site could embed instructions to manipulate the agent's behavior during the configuration generation process.
Ingestion points: HTML content retrieved from site paths such as usercp.php and torrents.php, or local HTML files provided by the user.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the processed HTML.
Capability inventory: The agent has the ability to write to local configuration files (assets/sites/{id}.json) and invoke other skills (Add Site).
Sanitization: No explicit sanitization or validation of the ingested HTML content is described before processing.
[DATA_EXFILTRATION]: The skill explicitly requests and uses session cookies to perform authenticated network requests. While this is necessary for its functionality (crawling private tracker sites), handling raw session tokens carries the risk of exposure if the agent is redirected to malicious domains or if the tokens are logged insecurely.
[COMMAND_EXECUTION]: The workflow requires the agent to execute network operations (crawling) and file system operations (reading manifests and writing site-specific JSON configuration files). These operations are used to manage the project's internal data structure.

nexusphpweb-site-config