The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external, third-party sources including YouTube channels, the Apple App Store, and general websites via scraping or RSS. This retrieved content is summarized and used to update 'Playbooks', which are then loaded back into the agent's context. This pattern represents an indirect prompt injection surface.
Ingestion points: External URLs processed via tools like manage_source, evaluate_url, or CLI commands like releases admin source create.
Boundary markers: The instructions do not specify explicit delimiters for untrusted content, though they mention processing steps.
Capability inventory: The agent can execute administrative commands (releases admin), create/delete sources, and update playbook notes.
Sanitization: The skill mentions 'summarizer-cleaned' and 'marketing-filtered' processing, which acts as a partial sanitization layer.
[EXTERNAL_DOWNLOADS]: The skill involves fetching content from remote URLs for evaluation and release tracking. While these are core functionalities, the retrieval of arbitrary external data is a security-relevant behavior.
[COMMAND_EXECUTION]: The skill relies extensively on a custom CLI tool (releases) and various administrative actions (e.g., releases admin org link) to manage the underlying system's state.

managing-sources