The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the coval CLI tool to list runs and simulations. This is standard functionality for interacting with the service provided by the skill author.
[EXTERNAL_DOWNLOADS]: Retrieves simulation transcripts from the official vendor API at api.coval.dev using curl. As this targets the vendor's own domain, it is considered a safe and expected operation.
[DATA_EXFILTRATION]: Uses the $COVAL_API_KEY environment variable to authenticate requests to the Coval API. This follows standard security practices for managing API credentials without hardcoding secrets.
[PROMPT_INJECTION]: The skill specifically instructs the agent to analyze transcripts for adversarial attack vectors like prompt injection and system-prompt extraction. It includes guardrails to ensure the agent reports these findings as failures rather than executing any instructions found within the untrusted data.

analyze-adversarial-report