The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input in the form of MySQL-style formulas and interpolates them into JSON payloads for shell commands. This creates a surface where a malicious user could attempt to craft a formula that injects additional JSON fields or manipulates the command structure.
Ingestion points: User-provided {formula} and {fieldName} in SKILL.md; external dataset schema fetched from the Domo API in Step 2.
Boundary markers: The skill does not use specific delimiters or instructions to treat user-provided formulas as data rather than instructions, nor does it provide guidance on handling potentially malicious content within these fields.
Capability inventory: The skill generates curl commands for multiple Domo API endpoints (including POST and PUT methods) that can modify datasets and card configurations.
Sanitization: The workflow includes a critical mitigation step requiring the agent to use the /api/query/v1/functions/validateFormulas endpoint to verify syntax and column references before generating any state-changing creation commands.
[COMMAND_EXECUTION]: The skill's primary function is to generate and output shell commands (curl) for the user. These commands are constructed using sensitive parameters like {devToken} and {instanceUrl}, which are intended for use with the official Domo API infrastructure.

beast-mode-creation