analyze-audio-quality-report
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill instructions include explicit security guardrails that prohibit the agent from exposing API keys, secrets, or private customer data. It also enforces evidence-based reasoning by requiring specific metric data or traces before making claims.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data including Coval report URLs, run results, and transcripts. While this involves ingesting untrusted content, the skill is focused on diagnostic analysis and does not grant high-privilege capabilities or execute arbitrary commands based on the data.
- Ingestion points: Processes report URLs, run IDs, transcripts, and simulation recordings as specified in the Inputs section.
- Boundary markers: No specific delimiters for external content are explicitly required in the prompt structure.
- Capability inventory: The skill utilizes the Coval CLI/API (vendor resources) for inspecting data and generates a structured analysis report in Markdown.
- Sanitization: The skill relies on the agent's internal reasoning and manual human review labels (when available) rather than automated processing of executable instructions within the transcripts.
- [SAFE]: The skill references 'Coval CLI/API', 'Coval report URLs', and 'run IDs', which are identified as legitimate resources belonging to the vendor 'coval-ai'. No evidence of typosquatting or impersonation of external services was found.
Audit Metadata