meeting-prep
Warn
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill constructs bash scripts by substituting user-provided participant and company names into double-quoted variable assignments, such as
NAME="[PARTICIPANT_NAME]". In a shell environment, content inside double quotes is subject to command substitution. If a user provides a name containing shell metacharacters like$(...)or backticks, thebashtool will execute those commands during script execution. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by searching and reading markdown files from the user's Obsidian vault. This content is then used to generate a meeting brief without appropriate safeguards.
- Ingestion points: The skill uses
grepandheadto pull text from arbitrary files in the vault (SKILL.mdSteps 1-3). - Boundary markers: The skill does not employ delimiters or system instructions to ignore potential commands embedded within the vault data.
- Capability inventory: The agent has
bashtool access, which allows it to read sensitive files and write new files to the filesystem. - Sanitization: There is no evidence of filtering or escaping logic applied to the content retrieved from the vault before it is processed by the agent.
Audit Metadata