Security audit
Security checks for vulnerabilities and agentic risk
This podcast-production skill is purpose-aligned and disclosed, with local episode/sponsor tracking that users should treat as potentially sensitive.
Install only if you are comfortable keeping podcast production records locally in podcast-data.json. Avoid saving confidential transcripts, sponsor contacts, or ad scripts unless you want them retained, and review or delete that file when the project no longer needs the history.
The README says the skill 'detects mode from context automatically' without defining strong boundaries between interview prep, transcript analysis, solo planning, and tracking actions. Ambiguous auto-routing increases the chance that the agent will infer the wrong mode and process sensitive user content or modify persistent episode history without an explicit, informed request.
The skill explicitly states it should activate on broad, casual phrases like 'I need show notes' or 'I have a guest this week,' which creates a real risk of unintended invocation. In an agent environment, this can cause the model to switch into a powerful workflow unexpectedly, leading to unnecessary data collection, transcript processing, web research, or persistent logging when the user did not clearly request the skill.
The skill advertises very broad trigger phrases such as ordinary podcast-related statements, which can cause the skill to activate in situations where the user did not intend to invoke persistent episode-management behavior. In this skill, unintended activation is more concerning because the capability includes reading and writing podcast-data.json, so accidental invocation can expose or modify stored transcript, sponsor, or episode metadata.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
You are a podcast production assistant helping creators eliminate the most tedious parts of producing an episode. You handle the work before the mic turns on and after the recording stops -- so the host can focus on the conversation itself.
You operate in three modes. Detect the mode from context -- don't ask the user to pick one.
---
The skill persists episode history, transcripts, sponsor relationships, and notes to podcast-data.json, but the user-facing description does not clearly warn that this data may be stored. This creates a privacy and consent issue because users may provide sensitive transcripts, contact details, or sponsorship information without realizing it can be retained across sessions.
The instruction to 'Detect the mode from context -- don't ask the user to pick one' leaves unclear how much evidence is required before selecting among multiple modes. Without clearer activation boundaries, similar user statements could trigger the wrong workflow unintentionally.
No suspicious patterns detected.