The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves data from external sources and instructs the agent to summarize the headlines. This pattern creates a surface where malicious instructions embedded in an external feed could potentially influence the agent's behavior.
Ingestion points: The scripts watch_rss.py, watch_http_json.py, and watch_github.py fetch content from user-provided URLs.
Capability inventory: The skill has the ability to perform network requests and write state files to the local filesystem.
Boundary markers: Content is formatted into Markdown sections, but there are no explicit instructions to the agent to ignore potential commands within the external data.
Sanitization: The scripts truncate content to a specific length but do not filter for natural language instruction patterns.
[COMMAND_EXECUTION]: The skill operates by executing Python scripts via the terminal and explicitly provides instructions for the agent to set up cron jobs to automate these executions for periodic monitoring.

watchers