qmd
Warn
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the user to install an external CLI tool globally using
bun install -g https://github.com/tobi/qmd. This executes code fetched directly from a third-party GitHub repository. - [PERSISTENCE]: The maintenance section provides specific instructions to add
qmd updateandqmd embedto the system'scronschedule. While intended for index freshness, this establishes persistent, automated execution of the tool on the user's machine. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process local Markdown collections which may contain untrusted content.
- Ingestion points: Files added via
qmd collection add(SKILL.md). - Boundary markers: None identified; retrieved content is interpolated into the agent context without explicit delimiters or warnings to ignore embedded instructions.
- Capability inventory: The tool executes various shell commands (
qmd search,qmd get,qmd query) to retrieve and display file contents (SKILL.md). - Sanitization: No sanitization or validation of the indexed markdown content is mentioned, creating a surface where malicious instructions in notes could influence agent behavior.
- [EXTERNAL_DOWNLOADS]: The skill mentions that GGUF models are automatically downloaded from remote sources on the first run to support semantic search features.
Audit Metadata