The Agent Skills Directory

[PERSISTENCE]: The installation instructions in instruct.md contain commands that modify the user's shell profile to maintain access across sessions.\n
Evidence: Step 5 in instruct.md instructs the agent to execute echo 'export QVERIS_API_KEY="<USER_PROVIDED_KEY>"' >> ~/.bashrc. This persists the environment variable in the user's bash configuration, despite the prose text in Step 3 claiming that modification of shell startup files should be avoided by default.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external API tools which could contain malicious instructions designed to influence the agent's behavior.\n
Ingestion points: Data is ingested from external APIs via the QVeris engine through scripts/qveris_client.mjs and processed via scripts/qveris_tool.mjs.\n
Boundary markers: Absent. The instructions in SKILL.md do not specify the use of delimiters or "ignore" instructions for external tool outputs.\n
Capability inventory: The skill has outbound network access to qveris.ai and api.qveris.cloud (via scripts/qveris_client.mjs) and the ability to execute shell commands to run the Node.js scripts.\n
Sanitization: Partial. The scripts/qveris_sanitize.mjs module provides sanitization of provider metadata (e.g., redacting internal routes and URLs), but does not filter for natural language instructions or prompt injection patterns in the returned data.\n- [EXTERNAL_DOWNLOADS]: The skill downloads its own implementation scripts and definition files from a remote server during the installation process.\n
Evidence: instruct.md uses curl and Invoke-WebRequest to fetch SKILL.md, qveris_tool.mjs, qveris_client.mjs, and qveris_env.mjs from https://qveris.ai/skill/. These resources originate from the vendor's own infrastructure.

qveris-official