The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches data from external, potentially untrusted websites using web extraction tools to compare prices against user-defined thresholds. This creates a surface for indirect prompt injection if the monitored web pages contain malicious instructions designed to influence the agent's behavior during the normalization or alerting phases.
Ingestion points: External URLs provided by the user are processed in Step 4 ("Fetch and normalize") using web_extract or browser_navigate.
Boundary markers: The instructions do not define explicit delimiters or "ignore instructions" warnings to separate external web content from the agent's internal logic.
Capability inventory: The skill has the capability to write state files to ~/.hermes/price-watches/, create and manage cronjob tasks, and navigate the web.
Sanitization: While the skill instructs the agent to extract specific data points (price, taxes, availability) and exclude volatile metadata, it lacks specific sanitization or escaping instructions for the content retrieved from external sources.
[PERSISTENCE]: The skill implements a persistence mechanism by creating recurring tasks to monitor prices across sessions.
Evidence: Step 3 explicitly instructs the creation of a job using the cronjob tool (e.g., cronjob(action="create", schedule="every 6h", ...)). This is the primary intended purpose of the skill and uses a high-level tool rather than low-level system configuration files, but it constitutes a persistent presence on the host system.

product-price-monitor