The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions advise the agent to insert user-provided strings (names and context) directly into JavaScript variable assignments within execution blocks.
Ingestion points: User-supplied search terms and context descriptions are used to populate the NAME and CONTEXT variables in SKILL.md.
Boundary markers: The skill relies on JavaScript string quotes but does not provide instructions to escape or validate user content.
Capability inventory: The arcmira_execute_read tool runs the resulting code with access to the arcmira API functionality.
Sanitization: No explicit sanitization or filtering logic is provided to clean user input before it is used in code generation.
[DYNAMIC_EXECUTION]: The skill uses the arcmira_execute_read tool to run JavaScript code generated from templates at runtime, which is its primary method for interacting with the Arcmira API.

sponsor-research