Security audit
Security checks for vulnerabilities and agentic risk
This podcast skill stores and summarizes podcast-related information locally in a way that matches its stated purpose, with some broad triggers users should treat carefully.
Before installing, be aware that the skill is designed to keep persistent local podcast records in ~/podcasts/. Review or clear that folder if you do not want subscriptions, listening interests, summaries, guest watchlists, or learning notes retained, and use explicit commands when adding shows or generating briefings.
The trigger 'User mentions a podcast → add to subscriptions' is overly broad because a casual mention does not clearly express consent to persistently store that podcast in the user's subscriptions. In this skill, the action writes to local storage under ~/podcasts/, so an incidental reference could cause unwanted data retention and inaccurate subscription state.
The phrase 'User asks for briefing → generate summary without them listening' is ambiguous because 'briefing' may be interpreted loosely and could trigger summarization when the user did not intend content processing or local summary generation. Given the skill stores generated briefings on disk, accidental activation can create unnecessary persistent records of listening interests and derived content.
The workspace section defines persistent storage for subscriptions, queue state, briefings, knowledge extraction, and guest watchlists, but the skill description does not warn users that these podcast activities and summaries are written locally. This omission undermines informed consent and can expose sensitive interest profiles, learning notes, or guest tracking data to other local users or tools.
The trigger phrase "what's this episode about?" is broad natural language that overlaps with common user speech, which can cause the skill to activate unexpectedly outside a clearly scoped command pattern. In an agent environment, overly generic triggers increase the risk of accidental invocation, context confusion, and unintended processing of user content.
No suspicious patterns detected.
