Security audit
Security checks for vulnerabilities and agentic risk
The skill is a coherent YouTube/TranscriptAPI integration, but its setup asks the agent to create or verify accounts and persist credentials broadly across future sessions.
Install only if you are comfortable letting the agent use TranscriptAPI for YouTube-related work and store a TranscriptAPI key for future sessions. Prefer creating the account yourself, providing the key through an approved secret mechanism, and asking the agent before sending sensitive research queries or private links to the service.
The guide directs the agent to persistently store an API secret across sessions and shells, creating long-lived credential access in the agent environment. For a YouTube-focused skill, this is an unjustified expansion of privilege and increases the blast radius if the agent, runtime, logs, or related skills are compromised.
The skill instructs the agent to perform third-party account registration, receive an email OTP from the user, and complete verification on the user's behalf. That is outside the declared YouTube research/transcript scope and grants the skill a credential-brokering capability that could be abused to create, verify, or bind external accounts without clear necessity or strong consent boundaries.
The invocation description is extremely broad ('use when YouTube is or could be relevant — even if not mentioned'), which can cause the agent to route many ordinary research prompts to this skill unnecessarily. That increases unnecessary third-party data exposure and the chance of external calls being made with user queries or identifiers when a local or less-privileged capability would have sufficed.
The skill instructs sending user-provided queries, video URLs, handles, and other identifiers to TranscriptAPI.com but does not prominently warn that this data leaves the agent environment and is shared with a third-party service. In combination with the broad routing language, users may have their research topics or linked content transmitted externally without clear notice or consent.
The instructions tell the agent to inspect runtime documentation, config files, and environment-secret mechanisms to determine how to persist credentials. This expands the skill into agent-environment reconnaissance and secret-management behavior beyond its stated function, which can expose internal configuration details or normalize overbroad access patterns.
The flow asks the user to provide sensitive data including an API key, email address, and OTP, then transmits and stores resulting credentials without a clear user-facing warning about sensitivity, retention, or storage implications. This weakens informed consent and raises the risk of users oversharing secrets into an agent workflow that may log or persist them.
No suspicious patterns detected.
