The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute an installation script directly from the vendor's GitHub repository using a curl | sh pattern. While this is used for installing the required veris-proxy tool, executing remote scripts is a high-risk operation. The skill correctly instructs the agent to ask the user for permission before proceeding.\n- [COMMAND_EXECUTION]: The skill involves complex shell command execution, including managing Docker containers and the veris-proxy binary. It uses dynamic command assembly, such as 'reconstructing' commands from header comments in a Dockerfile.veris file. To mitigate risks, the skill explicitly warns against directly executing content from these files due to the risk of malicious shell code hidden in comments.\n- [EXTERNAL_DOWNLOADS]: The skill fetches an installation script from raw.githubusercontent.com/veris-ai/veris-proxy/main/scripts/install.sh and pulls Docker images from a private Google Cloud registry (us-central1-docker.pkg.dev). These are identified as vendor-owned resources necessary for the skill's functionality.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from the repository and external documentation.\n
Ingestion points: Dockerfile.veris content, MCP-based testing guides, and service manuals via HTTP.\n
Boundary markers: Explicit instructions to 'never paste-execute' and to 'reconstruct' commands manually are included to mitigate the risk of embedded malicious instructions.\n
Capability inventory: Access to Docker CLI, veris-proxy tool, and shell command execution.\n
Sanitization: The skill relies on agent-level verification and manual command reconstruction instead of direct execution of untrusted snippets.

integration-testing