The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill communicates with the Vanta API (https://api.vanta.com) to fetch and update vulnerability status. This is a core part of its functionality and targets a well-known security service.
[COMMAND_EXECUTION]: The skill uses various shell commands (git, docker, gh, bundle, npm, pnpm, go) to manage repositories, build containers, and update dependencies. These are standard development tools and the usage is scoped to the skill's purpose of automated patching.
[DATA_EXPOSURE_AND_EXFILTRATION]: The helper script vanta.py reads credentials from ~/.vanta-credentials.json. While it accesses a sensitive file, it uses these credentials solely to authenticate with the Vanta API for the skill's primary function. It does not exfiltrate this data to unauthorized third parties.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Vanta API (CVE descriptions, package names). However, this data is processed as structured JSON and used as arguments for well-defined commands (like bundle update or apt install), minimizing the risk of prompt injection affecting the agent's logic. The output to Slack is also a summary of these structured fields.

patch-vulnerabilities