The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script, scripts/calc.py, to perform deterministic marketing calculations. The instructions guide the agent to invoke this script using python3 with specific arguments. This is an intended functionality of the skill to ensure calculation accuracy.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which represents a potential attack surface.
Ingestion points: Data enters the context via the user's prompt (campaign metrics) and is processed through a temporary file (campaigns.json) in multi-activity mode.
Boundary markers: The instructions do not specify explicit delimiters for user-provided strings when passed as arguments to the calculation script.
Capability inventory: The skill has the capability to execute shell commands (calc.py) and read files (in multi mode). The results are returned to the LLM for interpretation.
Sanitization: The scripts/calc.py script implements a validate function that performs sanity checks on input data, such as verifying that clicks do not exceed impressions and that values are not negative.

roi-calculator