The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses an 'operating prompt' in references/operating_prompt.md designed to create a blunt, Marc Andreessen-inspired persona. This prompt includes instructions such as "Do not provide disclaimers," "Do not inform me about morals and ethics," and "Never apologize for disagreeing." While these resemble patterns used in prompt injection to bypass safety guardrails, they are scoped here to the agent's tone and persona for business advice.
[DATA_EXFILTRATION]: The script scripts/anti_todo_card.py performs local file system operations, creating and managing a directory at ~/.andreessen-cards/ to store user task data. No network functionality is present in any of the skill's files, and no data is transmitted externally.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data, specifically task names in scripts/anti_todo_card.py and answers to strategic questions in assets/forcing_question_worksheet.md. This content is later echoed in summaries and verdicts, creating a potential indirect injection surface.
Ingestion points: Command-line arguments for tasks (--must-do, --did) and user responses during the worksheet interrogation workflow described in SKILL.md.
Boundary markers: None identified; user input is interpolated directly into responses.
Capability inventory: Local file system access (read/write JSON) and command execution for the provided Python scripts.
Sanitization: The Python scripts do not perform explicit validation or escaping of the user-provided string inputs.

andreessen