The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied vocabulary without specific boundary markers or instructions to treat the input as untrusted data. This could potentially allow a user to inject instructions disguised as vocabulary words.
Ingestion points: The skill instructions in SKILL.md direct the agent to process vocabulary words provided by the user.
Boundary markers: There are no delimiters (e.g., XML tags or triple quotes) defined to encapsulate user input.
Capability inventory: The skill frontmatter does not limit the agent's default toolset (e.g., shell access or file system tools), meaning an injection could theoretically target these capabilities if available to the agent. No additional scripts or tools are provided by the skill itself.
Sanitization: No input validation or filtering logic is specified.

mnemonics