The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the @tobilu/qmd package from the npm registry and automatically downloads approximately 2GB of local GGUF models from external sources during the first run.\n- [REMOTE_CODE_EXECUTION]: Fetches and executes a Node.js setup script from deb.nodesource.com using a shell pipe (curl | bash). This targets a well-known service for development environment configuration.\n- [COMMAND_EXECUTION]: The skill uses the terminal() tool to execute qmd, npm, brew, systemctl, and launchctl commands to manage search collections and services.\n- [DATA_EXFILTRATION]: The skill accesses sensitive local directories, including meeting transcripts, personal notes, and project documentation, to build its search index and retrieve document content.\n- [PRIVILEGE_ESCALATION]: Utilizes sudo for installing Node.js and system-level packages on Linux systems.\n- [PERSISTENCE]: Provides instructions to configure background services to maintain the search daemon across sessions using launchd on macOS (creating a .plist file in LaunchAgents) and systemd on Linux (creating a user service). This persistence is intended to keep the search models loaded in memory.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection.\n
Ingestion points: The skill retrieves content from potentially untrusted local files, notes, and meeting transcripts using qmd search and qmd get commands.\n
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions are provided in the prompt templates.\n
Capability inventory: The agent can execute shell commands via terminal() and use MCP tools to interact with searched data.\n
Sanitization: No sanitization or filtering of the retrieved document content is performed before it is added to the agent's context.

qmd