Security audit
Security checks for vulnerabilities and agentic risk
This is a documentation-only skill for using Gladia's remote transcription API, with expected upload, callback, and job-management behavior but limited privacy warnings.
Before installing, confirm you are comfortable sending selected recordings or media URLs to Gladia and, when using callbacks, only use webhook endpoints you control and trust. Be careful with confidential, regulated, or personally identifying audio; enable redaction or avoid optional analysis features when not needed.
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).
client.prerecorded().delete(job_id)
Raw REST: `DELETE /v2/pre-recorded/:id`
The skill explicitly encourages sending local audio files and external URLs to Gladia's remote transcription service but does not clearly warn that the content leaves the local environment and may contain sensitive data such as conversations, PII, or regulated information. In an agent setting, this omission can cause users or downstream automation to transmit confidential media to a third party without informed consent or proper policy review.
The documentation encourages use of a callback_url for async delivery but does not clearly warn that full transcription results, which may include sensitive audio-derived content such as PII, summaries, entities, and subtitles, will be transmitted to the specified external endpoint. In this skill’s context, that omission is security-relevant because the feature is explicitly designed for processing potentially sensitive recorded audio, making accidental exfiltration to an unintended or weakly protected webhook more likely.
No suspicious patterns detected.