The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a local Python script, panel.py, to manage the simulation logic, persona generation, and result tallying. This script is executed using the standard python3 interpreter and is restricted to the skill's local directory.
[INDIRECT_PROMPT_INJECTION]: The skill constructs prompts for sub-agent tasks by interpolating content from customer.json and pitch.md. This creates a surface for indirect prompt injection if the source data contains malicious instructions meant to influence the sub-agents.
Ingestion points: Content from founder/customer.json and founder/pitch.md is read and included in the sub-agent briefs generated by panel.py.
Boundary markers: The BRIEF template in panel.py utilizes explicit delimiters (=== WHO YOU ARE ===, === END ===) to isolate persona and pitch data from the sub-agent's instructions.
Capability inventory: Sub-agents are instructed to use their file-writing tool to save responses as JSON files in a dedicated answers directory.
Sanitization: The script performs JSON structure and schema validation on the sub-agent responses before saving them, although it does not sanitize the input content interpolated into the prompts.

founder-consumer