The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-supplied data (skill names and descriptions) which is then used to dynamically generate file system structures (directories and SKILL.md files) and Slack messages. This creates a surface where malicious user input could attempt to influence the agent's behavior or inject instructions into the generated skill metadata or team communication channels.
Ingestion points: The skill ingests user input for "skill name" and "description" during the initialization phase.
Boundary markers: There is no mention of delimiters, escaping, or "ignore embedded instructions" warnings for the user-supplied content within the generated templates or messages.
Capability inventory: The skill performs file system write operations (creating directories and SKILL.md files), local packaging (ZIP creation), and network communication via the Slack integration (Rube).
Sanitization: The documentation mentions validation for format and completeness but does not detail sanitization or filtering of the user-provided text for prompt injection patterns.
[COMMAND_EXECUTION]: The skill performs file system operations to create standardized directory structures (scripts/, references/, assets/) and files. While no code is provided, these operations represent a capability to modify the local environment.

skill-share